Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
Sources of Revenue and Local Government Performance:
Evidence from Colombia
Luis R. Martınez*
London School of Economics and Political Science
JOB MARKET PAPER
November 19 2015
Abstract
If government revenue is not coming out of their pockets, voters may be uninformed about itor uninterested in what happens to it, contributing to low accountability and poor governance.The present paper provides empirical evidence in support of this hypothesis by comparing theeffects of increases in internally-raised tax revenue and in royalties from the extraction of oilon local public good provision in a panel of Colombian municipalities. I find that an increasein property tax revenue, occurring as a result of an exogenous cadastral update, has a positiveeffect on several basic public services in the areas of education, health and water. These effectsare at least ten times larger than the effects of an equivalent increase in oil royalties, obtainedas a consequence of higher oil prices. The effect of oil royalties on public goods is basicallyzero, even though royalties are earmarked for this purpose. Using novel data on disciplinaryprocesses, I find that additional oil royalties increase the probability that local public officialsare prosecuted, found guilty, and removed from office, while increases in tax revenue have anegative effect on the likelihood of these events. These results suggest that the improvements inaccountability resulting from the devolution of expenditure responsibilities to local governmentsare reduced as the dependence on external sources of revenue, such as natural resource rentsor intra-government transfers, increases.
Visit http://sites.google.com/site/lrmartineza for the latest version.
*[email protected]. Department of Economics and STICERD. I would like to thank Gerard Padro i Miquel for hissupport and advice throughout this project. I am also grateful to Fernando Aragon, Oriana Bandiera, Pranab Bardhan, TimBesley, Florian Blum, Gharad Bryan, Francesco Caselli, Jonathan Colmer, Andrew Ellis, Miguel Espinosa, Claudio Ferraz,Greg Fischer, Lucie Gadenne, Maitreesh Ghatak, Anders Jensen, Camille Landais, Gilat Levy, Marıa del Pilar Lopez, RogerMyerson, Imran Rasul, Nelson Ruiz, Fabio Sanchez, Johannes Spinnewijn, Munir Squires, Guo Xu and to seminar/conferenceparticipants at LSE, EDEPO (UCL), PACDEV 2015, LACEA-PEG 2015, ZEW PF 2015, RES Jr. 2015, IIPF 2015, EEA 2015,OXDEV 2015, Frontiers of Urban Economics 2015 and NEUDC 2015 for comments and suggestions. I would also like to thankIvo Bischoff, Massimiliano Piacenza, Carlos Scartascini, Ju Qiu and Xiao Yu Wang for their thoughtful discussions. I thankMario Martinez at Instituto Geografico Agustın Codazzi for answering all my questions on cadastral updating and I also thankAdriana Camacho, Juan Camilo Herrera, Oskar Nupia, Monica Pachon, Fabio Sanchez, Rafael Santos and Ana Marıa Tribınfor generously sharing data with me. All remaining errors are mine.
1. Introduction
Inadequate provision of public goods is an obstacle to development in most low-income
countries (World Bank, 2004; Besley and Ghatak, 2006). A frequent way of addressing this
problem in recent decades has been through the devolution of expenditure responsibilities
to local governments (Gadenne and Singhal, 2014). These reforms have tried to exploit the
increased accountability of local public officials and they have had widespread support from
international organizations such as the World Bank (2000).1 However, this wave of decentra-
lization has met with only limited success so far, despite the fact that local democracy has
been found to provide strong incentives for good governance.2
It is also true that local governments in developing countries depend to a large extent
on external sources of revenue, such as transfers and natural resource rents. Several well-
identified studies have documented how increases in revenue from these sources appear to
have a very low impact on public good provision, often leading to a worsening of corruption
instead.3 It thus seems plausible that the way in which local public finances are organized is
contributing to the suboptimal provision of local public goods across the developing world. If
revenue is not coming out of their pockets, voters may be uninformed about it or uninterested
in what happens to it, failing to hold the government accountable as a result.4
In the present paper, I test the hypothesis that internal revenue, raised through local
taxation, has a larger effect on the provision of public goods than revenue from an external
source. I compare the effects of increases in local tax revenue and in royalties from the
extraction of oil on the provision of public goods in a panel of Colombian municipalities. I
show that internally-raised tax revenue has a much larger impact on public good provision
than revenue from oil royalties. Using novel data on disciplinary offences, I further show that
this difference is driven by the opposite effects of revenue from the two sources under study
on the misbehavior of local politicians.
A comparison of this nature faces several challenges. We must first find a setting where lo-
cal governments provide public goods using revenue from both internal and external sources.
1In the words of Bardhan (2002, p.185), “In matters of governance, decentralization is the rage.”2See Faguet (2014) and Mookherjee (2015) for recent reviews on decentralization. For evidence on local
democracy in developing countries, see Ferraz and Finan (2011); De Janvry et al. (2012); Martınez-Bravoet al. (2014); Fujiwara (2015).
3See Fisman and Gatti (2002); Reinikka and Svensson (2004); Vicente (2010); Caselli and Michaels(2013); Brollo et al. (2013); Litschig and Morrison (2013); Maldonado (2014); Olsson and Valsecchi (2014).
4The idea that taxation improves governance is not new. It can be found in comparative papers on thedevelopment of modern Europe (North and Weingast, 1989) or on the ‘rentier states’ of the Middle East(Mahdavy, 1970; Beblawi, 1990; Ross, 2001). In development economics, this idea is present in discussionson foreign aid (Bauer, 1972; Easterly, 2006; Collier, 2006) and on state capacity (Besley and Persson, 2011,2013, 2014). In public economics, it is at the core of the ‘second generation’ approach to fiscal federalism(Oates, 2005; Weingast, 2009) and it is related to the idea of ‘fiscal illusion’ (Dollery and Worthington, 1996).
1
We must also have access to plausible sources of variation not only in external revenue, which
the previous literature has accomplished with some success, but also in internally-raised tax
revenue, which is a more daunting task and has seldom been done before. Finally, our com-
parison must account for the fact that revenue can be used for many different purposes, so
a careful choice of outcomes is needed.
Colombian municipalities are responsible for the provision of basic public services and
finance them with a mix of local taxes, transfers from the central government and royalties
from the extraction of natural resources.5 The royalties received by resource-producing muni-
cipalities are formula-determined and amount to a fixed share of the market value (at world
prices) of the extracted resources. Oil is the most important source of royalties in Colombia
but the country is a small player in the oil market and is unable to affect world prices.6
Hence, I exploit time variation in the world price of oil between 2005 and 2011, together
with cross-sectional variation in oil intensity (using average municipal oil royalties between
2000 and 2004) to estimate the effect of royalties on local public goods.
The municipal expenditure of natural resource royalties is heavily regulated. At least
75 % of royalties must be spent on five specific indicators in the areas of education, health,
drinking water and sanitation until targets are met. Therefore, these indicators are the best
places to look for the impact of royalties on public services. Additionally, the targets are
closely related to the attainment of the Millenium Development Goals and provide valuable
information on local living standards. The four indicators for which yearly data is available
are my main outcomes of interest: the net enrolment rate in basic education, the infant
mortality rate, the percentage of poor population with subsidized health insurance and a
water contamination index.
I compare the effects of oil royalties, an external source of revenue, on the indicators
above to the effects of property tax revenue, an internal source of revenue, on the same
set of indicators. The property tax is one of the main local taxes in Colombia. The base
of the tax is the value of the properties in the municipality’s official property register or
cadastre. Each year the national geography institute run by the central government updates
the cadastres of some municipalities. These updates mainly involve reassessing the value
of existing properties and they lead to a sharp increase in property tax revenue. I provide
evidence that the timing of the updates is determined by the geography institute and is
uncorrelated with variation in local political and economic characteristics, making these
updates a source of plausibly exogenous variation in property tax revenue. Furthermore,
5Colombia is divided into 1100 municipalities, each of which belongs to one of 32 departments, similarlyto US states and counties.
6In the appendix I provide very similar results for coal, which is the second most important source ofroyalties. Together, oil and coal account for over 90 % of royalties in the period 2005-2011.
2
since municipalities have discretion over the expenditure of the tax revenue, the comparison
of the effect of royalties and tax revenue on the outcomes on which only royalties have to be
spent is biased against the hypothesis I want to test .
Nevertheless, the estimates from an instrumental variables model with municipality and
department-year fixed effects indicate that an increase in property tax revenue has a positive
and statistically significant effect on educational enrolment and water quality. Additional
property tax revenue also increases the probability of achieving universal coverage of poor
population with subsidized health insurance. These effects are at least one order of magnitude
greater than (and statistically different from) the effects of an increase in royalties of the
same size. In fact, I find a striking result: the effect of additional royalties is basically zero
for all outcomes and the point estimates are negative in several cases.
One potential concern regarding these findings is that different types of municipality may
raise revenue from different sources. However, I show that the reported effects of cadastral
updating are not heterogeneous by oil intensity. Another potential problem is that differences
in expenditure could arise because cadastral updates lead to a stable increase in tax revenue
while oil price shocks lead to temporary fluctuations in royalties. To address this concern, I
show that $1 from either source translates into $1 of investment in infrastructure (gross fixed
capital formation), although only tax revenue leads to an increase in the number of schools.
A further alternative explanation for the findings is that royalties are spent on large-scale
projects with less immediate effects. I deal with this possibility by showing that the outcomes
of interest in oil-royalty recipients are almost identical in 2011 to what they looked like in
2005. Finally, I also provide evidence against the possibility that the results are affected by
potential violations of the exclusion restriction due to non-fiscal effects of oil-price shocks.
Given that the purchasing power of revenue is the same across sources, the reported
heterogeneity in the effects of internal and external revenue on public good provision must
be driven by endogenous differences in the response of local public officials to extra revenue
from these sources. To further explore if this is the case, I gather a new dataset on disciplinary
processes involving local public officials in Colombia. I find that an increase in oil royalties
leads to an increase in the probability that the municipal mayor and top members of staff are
prosecuted, found guilty and removed from office by a national watchdog agency. Increases
in property tax revenue, on the other hand, reduce the probability of these events.
These results are consistent with the idea that taxation makes voters either more able
or more willing to hold the government accountable (Paler, 2013). Taxation may provide
voters with information on public revenue or may raise their awareness about it. Taxation
also reduces private consumption and is generally disliked by voters, who may punish the
government that fails to compensate them. In the theoretical appendix, I provide a model
3
of political agency with career concerns that illustrates how both information-based and
preference-based mechanisms can explain the results from the empirical exercise.
The present paper’s main contribution is to the empirical literature studying the rela-
tionship between public finance and governance.7 Within this literature, two recent papers
have studied the differential effects of internal and external revenue on public good provi-
sion, with mixed findings. Borge et al. (2015) show that additional rents from hydro-power
production reduce the efficiency of public expenditure less than increases in other revenue in
Norwegian municipalities. In the most closely related contribution to the present paper, Ga-
denne (2015) reports improvements in educational infrastructure for Brazilian municipalities
that enroll in a tax modernization program, while higher transfers have no effect.
The main challenge that this line of research still faces is coming up with plausibly
exogenous sources of variation in tax revenue, as changes in tax bases and tax rates are likely
to be endogenous to political and economic factors that can potentially affect outcomes of
interest. The present paper contributes to this literature by exploiting the quasi-random
timing of cadastral updates as a source of variation in internal revenue.8 These updates
lead to an increase in tax revenue that is not correlated to changes in political or economic
conditions, nor in tax administration or structure. I observe cadastral updates for 60 % of
municipalities in a five-year period, which ensures the representativeness of the results among
Colombian municipalities. It also allows me to show that the benefits of taxation extend to
the same municipalities where oil royalties have no effect. The present paper also contributes
to the existing literature by providing evidence on the heterogeneous effects of internal and
external revenue on politicians’ misconduct, based on novel data from disciplinary processes.
This paper is also related to to the empirical literature that uses sub-national data to
study the effects of natural resource rents.9 The theoretical model I develop also complements
previous contributions on the political resource curse by exploring the heterogeneous political
effects of resource rents relative to tax revenue.10 The paper contributes as well to the ‘second
7Zhuravskaya (2000) documents the negative effects of transfer offsets to increases in tax revenue inRussian cities. Ross (2004) reports cross-country evidence on the link between taxation and democracy.Paler (2013) and Martin (2014) provide experimental evidence on people’s higher willingness to hold thegovernment accountable when they are taxed. Borge and Rattsø (2008) and Sanchez and Pachon (2013)show that property taxes improve the efficiency and amount of public services in Norway and Colombia,respectively. Casaburi and Troiano (2015) report positive effects on local governance from the cadastralregistration of properties in Italy.
8Sanchez and Pachon (2013) use an IV strategy based on cadastral depreciation and find that educationalenrolment and water quality improve in Colombian municipalities that collect more taxes. I build on theirwork by providing the necessary evidence on the exogeneity of the timing of cadastral updates. Additionally,while Sanchez and Pachon (2013) study the effect of fiscal capacity on public good provision, I answer adifferent question related to the heterogeneous effects of internal and external revenue.
9See Caselli and Michaels (2013); Maldonado (2014); Ferraz and Monteiro (2014); Olsson and Valsecchi(2014); Herrera (2014); Carreri and Dube (2015).
10See Caselli (2006); Mehlum et al. (2006); Robinson et al. (2006); Caselli and Cunningham (2009); Brollo
4
generation’ literature on fiscal federalism by providing evidence on the importance of local
fiscal incentives.11
The rest of this paper is organized as follows. Section 2 presents the conceptual framework
behind the main hypothesis to be tested. Section 3 provides background information on the
setting for the empirical exercise. Section 4 presents the data and discusses the empirical
strategy. The main results are shown in section 5, leaving robustness checks and extensions
for section 6. Section 7 concludes.
2. Conceptual Framework
In this section I explore the reasons why we might expect internal and external revenue to
have heterogeneous effects on governance. One straightforward reason is because voters are
better informed about changes in taxation than about changes in external revenue. In the
theoretical appendix I explore this mechanism in the context of a political agency model with
career concerns.12 In the model, voters receive a noisy signal on public revenue, the precision
of which is improved by the share of taxes in total revenue. As the revenue signal becomes
more precise, voters are more able to infer the incumbent’s ability after observing public
good provision. Hence, taxation makes the voters’ posterior beliefs on the incumbent’s ability
more sensitive to observed public goods and this leads to higher effort by the incumbent and
more public goods. As external revenue increases, on the other hand, voters become less
well informed about revenue and this has a negative effect on the incumbent’s effort. Thus,
external revenue has a smaller effect on public goods than tax revenue.13
This informational explanation is consistent with various findings from the empirical li-
terature. Recent studies provide evidence in support of the idea that voters find it difficult
to establish the contribution of government to observed outcomes (Leigh, 2009; De la O,
2013; Guiteras and Mobarak, 2014). Recent research also indicates that voters are relati-
vely uninformed about changes in external revenue (Reinikka and Svensson, 2004; Ferraz
and Monteiro, 2014; Gadenne, 2015). Additionally, there is a large literature showing that
governments are generally more accountable to voters that are better informed.14
et al. (2013); Matsen et al. (2015)11See Bardhan (2002); Oates (2005); Bardhan and Mookherjee (2006); Faguet and Sanchez (2008, 2014);
Weingast (2009). Glaeser (1996) and Hoxby (1999) explore the potential of the property tax to act as adisciplining device for local governments.
12The model is an extension of the canonical career concerns model of Persson and Tabellini (2000) thatincorporates ‘signal-jamming’ a la Holmstrom (1999). Alesina and Tabellini (2007) and Matsen et al. (2015)use similar extensions to answer very different questions.
13Gadenne (2015) develops a similar model of moral hazard.14See Besley and Burgess (2002); Reinikka and Svensson (2005); Ferraz and Finan (2008); Bjorkman and
Svensson (2009); Snyder and Stromberg (2010); Banerjee et al. (2011); Fergusson et al. (2013); Chong et al.
5
One may still wonder how can residents of resource-rich areas not be aware of the flow
of resource rents to their government. The point here is that even if voters in these areas
know about the abundance of natural resource rents, they must still pay close attention to
fluctuations in prices and output to be well informed about the change in these rents. This
is important because both the model and the empirical exercise to follow are concerned with
changes in revenue from different sources, rather than with their average level.
One could also wonder how informative it is to pay your own taxes in a world with
significant heterogeneity in tax liabilities. Although this does raise the question about which
are the taxes that matter, it is not a major concern for the empirical exercise on Colombia as
the cadastral updates that I study lead to a municipality-wide simultaneous increase in tax
liabilities. A related question is whether increases to taxation simply make voters, who are
already well informed about revenue, more aware of the public purse and its use (increased
salience). An explanation along these lines seems particularly plausible because the property
tax that I study in the empirical exercise stands out in this respect, as it is a yearly out-
of-pocket tax payment on an illiquid asset (Cabral and Hoxby, 2015). Additionally, there is
evidence that people’s response to taxation is affected by the salience of taxes (Chetty et al.,
2009; Finkelstein, 2009).
There are, of course, other channels besides information and salience through which the
source of revenue may affect accountability and government performance. It is possible,
for instance, that voters simply dislike taxation and punish the incumbent for it unless he
compensates them with improved public services. Martin (2014) develops a model along
these lines in which loss-averse voters derive utility from punishing a corrupt government.15
In the appendix, I present an alternative version of the theoretical model described abo-
ve in which the marginal utility of public goods is decreasing in private consumption and
voters can acquire costly information on public revenue. I show that taxation may improve
incumbent effort and public good provision, even if it is not by itself directly informative,
because it induces the acquisition of costly information on government revenue due to its
negative effect on disposable income.
There is some empirical evidence supporting these preference-based channels. Both Paler
(2013) and Martin (2014) find that participants in lab experiments are more willing to engage
in costly punishment of a misbehaving government when the source of revenue is taxation
than when it is external, even when information is held constant across treatments. There is
also a large literature on reciprocity that has found that people are willing to incur in costly
(2015).15In my model, forward-looking voters cannot credibly commit to vote against the incumbent if they
believe him to be of higher ability than his opponent in the election.
6
punishment of what they consider to be unfair behavior (Fehr and Gachter, 2000).
Overall, taxation may either increase citizens’ willingness to hold the government ac-
countable or their ability to do so (Paler, 2013). Although the available data does not allow
me to distinguish between the explanations discussed above, all of them predict that tax
revenue has a larger effect on public good provision than external revenue. In the empirical
exercise that follows I provide empirical evidence in support of this hypothesis. Additional
evidence from disciplinary processes indicates that the actions of local governments respond
differently to increases in local taxation than to increases in external revenue.
3. Background
Following decentralization reform in the early 1990s, Colombian departments and muni-
cipalities became jointly responsible for the provision of basic public services in the areas of
education, health, drinking water and sanitation. The central government provides funding
for related expenditures through a system of earmarked and formula-determined transfers
called “Sistema General de Participaciones” (SGP). These transfers account on average for
63 % of a municipality’s total revenue and must be kept in a separate account from other
local revenues. Autonomy over the expenditure of transfers and over the administration of
public services varies across municipalities and across sectors, with the specific responsibili-
ties of the different levels of government for public service provision being somewhat blurry
(Alesina et al., 2005).16 However, as will become clear below, I verify that all municipalities
in the sample are legally able to use revenue from the two sources under study to improve
the outcomes of interest.
Taxes are the second most important source of municipal revenue after transfers and
contribute on average with 44 % of current receipts and 13 % of total revenue. The main local
taxes (and their average shares of tax revenue) are the property tax (34 %), the business tax
(17 %) and the petrol surcharge (22 %).17 The property tax is the most important source
of tax revenue for slightly more than one half of municipalities, but its relative importance
decreases with population size.18
16After an additional reform in 2001, municipalities “certified” by the Ministries of Education or Healthstarted to directly manage the transfers earmarked for these areas (Cortes, 2010; Brutti, 2015). Otherwise,transfers are managed by the departmental government. Certified municipalities also have greater autonomyin the management of the local education and health systems. However, the provision of health services ishighly regulated, even for certified municipalities, and must take place through special firms called “EmpresasSociales del Estado” (ESE). In the case of water and sanitation, municipalities manage the share of transfersearmarked for this purpose unless they are “de-certified” by the Superintendent for Public Services.
17Other local taxes include those for car registration and for the display of billboards and banners.18Glaeser (2013) reports that local public finances in the US are not very different, with intra-government
transfers and property taxes being the most important sources of revenue for all but the largest cities.
7
Municipalities have discretion over the expenditure of property tax revenue, except for
a fixed share that must be transferred to the corresponding environmental agency.19 All
municipalities can supply funding for the provision of education and they can also invest in
infrastructure and school equipment. In health, all municipalities are responsible for provi-
ding subsidized insurance to the population classified as poor by the national government’s
proxy-means-testing targeting system (SISBEN). Tax revenue can also be used for public
health initiatives such as vaccination campaigns (vaccines are provided at zero cost by the
central government). Regarding water and sanitation, municipalities can invest property tax
revenue in infrastructure or can also use it to provide subsidized access to these services to
the poor.
The property tax is levied on the cadastral value of all real estate in the municipality. The
cadastre or land register is the official record of the physical and economic characteristics
of all properties in a municipality. The three largest cities (Bogota, Medellın and Cali) as
well as the department of Antioquia have their own cadastral agencies and are excluded
from my sample. All others (86 % of municipalities) are under the authority of the National
geography institute, Instituto Geografico Agustın Codazzi (IGAC), an agency run by the
central government. IGAC periodically updates the cadastres under its control. This mainly
involves reassessing existing properties but also, to a much lesser extent, including in the
cadastre previously unregistered properties.20
The third most important source of local revenue are royalties from the extraction of
natural resources. Royalties are paid by firms to the central government according to a set
of fixed resource-specific formulae of the form
royalty = output × world price (USD) × exchange rate (COP/USD) × royalty rate
The vast majority of this revenue is then transferred to producing municipalities and de-
partments, as well as to the port municipalities from where resources are shipped, according
to predetermined shares. The main source of royalties is the extraction of oil. Oil royalties
amounted to 3.5 billion USD between 2005 and 2011, which accounts for 69 % of all royalties
paid in this period.21 18 % of municipalities received oil royalties and these represented, on
average, 23 % of total revenue.
Gadenne and Singhal (2014) show that dependence on external revenue is greater for local governments indeveloping countries.
19There are 34 such agencies in the country. Some cover a handful of municipalities while others covermultiple departments. The percentage transferred must be between 15 % and 25 % of property tax revenue.
20See Iregui et al. (2003, 2004) and Sanchez and Espana (2013) for further information on the propertytax in Colombia and on cadastral updating.
21Royalties are also paid for the extraction of coal, precious metals, gemstones, iron, copper, nickel andsalt.
8
By law, at least 75 % of royalties must be spent on education, health, drinking water
and sanitation until specific targets are met for the specific set of indicators listed in Table
1. These are the net enrolment rate in basic education (years 1-9, ages 6-14), the infant
mortality rate, the percentage of poor population with subsidized health insurance and the
percentages of population with access to drinking water and sanitation. Regarding water, it
is only considered suitable for human consumption (“drinkable”) if it scores less than 5 in
a water contamination index ranging from 0 to 100. Once all targets are met municipalities
keep receiving royalties and can spend them on priority projects from the mayor’s government
plan.
Royalties can be used to finance the provision of education if SGP transfers are shown to
be insufficient. Otherwise, royalties can be spent on education infrastructure, school equip-
ment or transportation. Royalty recipients can reduce infant mortality through public health
policies or by setting up emergency health posts for common infant diseases. Royalties can be
used for expenditures related to water and sewerage projects, such as initial studies, designs
and construction.
A reform in 2012 significantly reduced the royalties received by producing municipalities
and modified expenditure procedures. I drop years after 2011 from the sample for this reason,
but also due to some data limitations.
The top municipal authority is the mayor. Mayoral elections take place simultaneously in
all municipalities every four years. The local legislative body is the municipal council, which
varies in size with population and is elected at the same time as the mayor. The council
has to approve both the mayor’s general government plan and the annual budget and it is
responsible for monitoring their execution.
The watchdog agency Procuradurıa General de la Nacion (PGN) oversees public emplo-
yees’ compliance with the relevant disciplinary norms. This includes local public officials,
such as the mayor, top members of staff (e.g. secretary of education) and municipal council
members. PGN may start an investigation based on news reports, tip-offs, audit results and
reports from other government agencies such as the fiscal watchdog Contralorıa General de
la Republica (CGR). PGN can hand out sanctions ranging from fines and short suspensions
for small offences, to the removal from office and a ban from future public employment and
office. These latter sanctions are reserved for serious offences, such as intentional violations
of procurement, hiring and electoral laws.
9
4. Empirical Strategy
When voters are more aware or more affected by the way in which their local government
collects revenue, government accountability and public service provision improves. To test
this hypothesis, I use panel data for Colombian municipalities between 2005 and 2011 and
I compare the effects on local public good provision of shocks to local tax revenue to the
effects of changes in an external source of revenue, the royalties from the extraction of oil. I
exploit the timing of cadastral updates and fluctuations in the world price of oil as sources
of plausibly exogenous variation in property tax revenue and royalties, respectively. I further
test for heterogeneous effects of these sources of revenue on government performance by
looking at disciplinary processes involving local public officials.
I explain this empirical exercise in the following sub-sections. First, I introduce the data
employed. I then present the outcomes of interest. Finally, I discuss the identification strategy.
4.1. Data
The empirical exercise described above requires three main pieces of data. First, I need
data on the different sources of revenue of municipal governments. Second, I require informa-
tion on local public goods. Finally, I must also have information on the sources of variation
of both internal and external revenue. This means having data on cadastral updates, on the
world price of oil and on local oil abundance. All of these data must be available at the
municipality-year level to allow me to control for unobserved heterogeneity across municipa-
lities and over time.
Data on municipal public finance comes from the yearly balance sheets reported by each
municipality to the Office of the Comptroller General for the purpose of fiscal control. These
balance sheets have disaggregated information on all sources of revenue, including tax re-
venue (by type of tax), transfers and royalties. Information on expenditure is also available
in these balance sheets, disaggregated between current expenditure (operating costs) and
investment. I express all money values in millions of 2004 Colombian Pesos per capita (un-
less otherwise stated) using the Consumer Price Index and population estimates from the
National Statistical Agency (DANE).
Data on local public goods comes from various sources: the net enrollment rate in basic
education and the infant mortality rate are provided by the Ministry of Education and
DANE, respectively.22 The percentage of poor population with access to subsidized health
22The net enrolment rate is calculated by dividing the number of children with ages 6 to 14 enrolledin school years 1 to 9 by the number of children in this age group. Since data on enrolment and data onpopulation come from different sources, the resulting figure can actually exceed 100 %. I censor enrolmentrates above 100 % but the results are robust to using the original data.
10
insurance comes from the Ministry of Health.23 The water contamination index, Indice de
Riesgo de la Calidad del Agua (IRCA), has been calculated by the National Health Institute
since 2007. The other indicators are available since 2005. In the following section I explain
why I choose these indicators as the main outcomes of interest.
Regarding cadastral updating, IGAC has yearly data on the number of properties, the
cadastral value and the year of the last cadastral updates for both the urban and rural areas
of each municipality under its supervision. As mentioned above, the three largest cities in
the country (Bogota, Medellın and Cali) as well as the department of Antioquia have their
own cadastral agencies and are excluded from the analysis,so the final panel includes 969
municipalities (out of a total of 1123) between 2005 and 2011. In the empirical exercise I do
not distinguish between urban and rural updates, but the results are robust to the exclusion
of rural updates.
Data on oil royalties comes from the National oil company, Ecopetrol, for the period
2000-2003 and from the National Hydrocarbons Agency, Agencia Nacional de Hidrocarburos
(ANH), for the period 2004-2011. I use the average petroleum spot price from the IMF’s
International Financial Statistics (IFS).
The other main piece of information is a novel dataset on disciplinary processes involving
local public officials carried out by the Inspector General’s Office, Procuradurıa General de
la Nacion (PGN). I constructed the dataset using news reports from PGN’s website for the
years 2004-2015. For each process, I recorded the names of the accused, their roles in the
public administration, the nature of the charges, the timing of the events, the stage of the
process and the outcome. I collapse this data to the municipality-local political term level
(2001-2003, 2004-2007 and 2008-2011), since it is not always possible to pin down at the
yearly level the timing of the events for which officials are being accused.
Summary statistics for all variables employed in the paper are provided in Table 2.
4.2. Local Public Goods
As mentioned above, Colombian law stipulates that at least 75 % of royalties must be
spent on the improvement of basic public services until targets are met for a specific set of
indicators. The targets are displayed in column 1 of Table 1. I also show in column 2 the
average value for each indicator in 2005, the first year for which data is available, among oil-
royalty recipients. At the start of the sample, the average municipality receiving oil royalties
does not meet any of the targets. Column 3 provides further evidence on the low levels
of target achievement. It shows that, with the exception of access to drinking water, the
23Poor is defined as belonging to categories 1 or 2 of the Colombian proxy-means-testing system SISBEN.
11
percentage of oil-royalty recipients reaching each target is less than or equal to 30 % in all
cases.
The low levels of compliance imply that the targets were binding constraints during the
sample period and that royalties had to be spent on the improvement of the indicators in
Table 1. Administrative data on the expenditure of royalties, which is only available for
recent years, confirms that between 2010 and 2011 the average oil-royalty recipient spent
approximately 80 % of royalties on the attainment of the targets. Although municipalities
could potentially move revenue from other sources to other purposes as royalties increase,
I find that the ratio of royalties to own revenues among oil-royalty recipients averages 1.58
during the sample period. This means that even with full crowding-out of own investment
we can be sure that royalties are spent at the margin on meeting the targets.
Hence, the indicators in Table 1 are the best places to look for the impact of additional
royalties and I use them as the main outcomes of interest for the empirical exercise. Another
reason to focus on these indicators is because they are also a valuable source of information
on local living conditions and the targets are closely related to the attainment of some of the
United Nations’ Millenium Development Goals (MDG), such as achieving universal primary
education and reducing child mortality.
This choice of outcomes does introduce a bias in the comparison I carry out, since muni-
cipalities have discretion over tax revenue and are not required to spend it on the outcomes
of interest. The effect of royalties should be mechanically large, when compared to the effect
of tax revenue, due to the higher required propensity to spend revenue from royalties on the
studied outcomes. However, this bias works against the hypothesis that I want to test.
Yearly data between 2005 and 2011 is only available at the municipality level for four of
the indicators in Table 1: the net enrolment rate in basic education, the infant mortality rate,
the percentage of poor population with access to subsidized health insurance and, starting in
2007, the index of water contamination. Data on the percentages of population with access to
drinking water and sewerage is only available in the 2005 population census, so I leave these
two indicators out of the analysis. However, as mentioned above, access to drinking water is
a composite indicator that requires residents to have access to water and this water to be
suitable for human consumption, which is measured by the included water contamination
index. I verify that there is a strong cross-sectional correlation between the values of the
two omitted indicators in 2005 and the baseline score from 2007 in the water contamination
index.24 Additionally, Sanchez and Vega (2014) report a large positive effect of access to
drinking water on infant mortality, so this latter indicator could also capture improvements
24Sanchez and Pachon (2013) find a positive cross-sectional effect of local taxation on access to drinkingwater in 2005.
12
in access to water and sanitation.
4.3. Identification Strategy
I exploit the availability of yearly data on the outcomes of interest and on revenue from
different sources for each municipality to estimate a model with municipality and department-
year fixed effects. This allows me to control for unobservable persistent heterogeneity across
municipalities as well as for common shocks affecting all municipalities simultaneously, with
this effect being potentially heterogeneous across departments. I thus identify the effects of
revenue from the two sources on the outcomes of interest from within-municipality variation
over time, net of common within-department time effects.
Still, OLS estimates of the fixed effects model could be affected by reverse causality
or by omitted variable bias. For example, a municipality could collect more taxes because
educational enrolment has risen due to the implementation of some other policy. Similarly,
royalties could increase due to the discovery of oil, which is likely to have a direct effect on
the outcomes of interest. Hence, I search for plausibly exogenous variation in both sources
of revenue in order to be able to claim that the estimates capture causal effects.
I use the timing of cadastral updates by IGAC as a source of exogenous variation in
property tax revenue. Colombian law requires municipal cadastres to be updated every five
years, but this condition is rarely satisfied. During the sample period, the average urban
update takes place 11.4 years after the previous one, with this number being slightly lar-
ger (12.7) for rural updates. I address potential concerns about the endogenous timing of
cadastral updates in three ways.
First, I check that the timing of the update is not affected by observable municipal
characteristics. I do this by estimating a series of bivariate regressions:{D(update)i,j,t+1 = αi + δj,t + βkX
ki,t + εi,j,t
}Kk=1
(1)
where Xki,t is a time-varying characteristic indexed by k and D(update)i,j,t+1 is a dummy
equal to one the year before the update comes into effect. I define the dependent variable
in this way to account for the fact that updates that take place in year t only come into
effect the first of January of year t + 1. I demean all variables using municipality (αi) and
department-year (δj,t) fixed effects to ensure that I look at the variation that I will exploit
in the main regressions.
Which are the observable characteristics that I study? I include the natural log of popu-
lation and the share of rural population to capture both between-municipality and within-
municipality (urban-rural) migration. I also study all other sources of revenue to check
13
whether updates try to offset or to complement other changes in revenue. I analyze poli-
tical competition, party affiliation and alignment across all levels of government: municipal
(mayor, council), departmental (governor) and national (president, congress). I look as well
at national policies, such as the number of families enrolled in the conditional cash-transfer
program Familias en Accion and the value of new loans made by the central government’s
agricultural bank. I check if cadastral updates are correlated with visits to the municipality
by President Alvaro Uribe because these visits led to significant policy commitments (Tribın,
2014). Finally, I look at indicators on crime, illegal armed group presence and cultivation
of narcotics. Although I am unable to rule out that variation in unobservables is affecting
the decision to update, it is not easy to think of changes in unobserved characteristics that
would not be picked up by the variables studied.
The results are presented in columns 1 and 2 of Table 3. Only the number of parties
participating in council elections, out of thirty variables considered, has a correlation with
the timing of cadastral updates that is statistically significant at the 10 % level.25 Although
this correlation could be spurious, I verify that results are robust to including any of the
variables considered as controls. I also show in the appendix that the results are robust
to the inclusion of municipality-term fixed effects, which capture any unobserved within-
municipality heterogeneity across local political terms. Furthermore, I check in columns 3
and 4 that the results are robust to adding separate dummies for the first five years after
urban and rural updates as controls. I ensure this way that the point estimates are not being
artificially attenuated by the fact that the probability of updating is essentially zero in the
years right after an update.
The second piece of evidence on the exogeneity of the cadastral updates is based on the
study of the supply of updates by IGAC. The sample period has the special feature that
IGAC received strong incentives to increase the number of updates. More specifically, Alvaro
Uribe included as part of his official government goals for his first term as President (2002-
2006) that IGAC should have the urban cadastres of all municipalities up to date (less than
five years old) by the end of the administration. For his second administration (2006-2010),
Uribe set as goals for IGAC to have 90 % of urban cadastres and 70 % of rural cadastres up
to date. Additionally, funding to IGAC increased, including an IDB loan for the purpose of
financing the cadastral updates of 145 municipalities in 2007.
Figure 1 shows the number of updated municipalities per year disaggregated by the type
of update. I indicate for each year who was President at the time, bearing in mind once
25Sanchez and Pachon (2013) and Sanchez and Espana (2013) report results from similar regressions usinga logit model. Although these authors find significant correlations with transfers, income and some politicalcharacteristics, the difference is probably driven by the lack of municipality fixed effects in those estimations.
14
again that there is a one-year lag in the validity of updated cadastres. The graph shows
that the number of updates increased sharply during the Uribe years: between 2006 and
2010, 60 % of municipalities have a cadastral update. This large number of updates coincides
with the positive supply shock caused by IGAC’s targets. Additional evidence that these
changes were driven by IGAC’s supply of updates comes from the fact that the share of
urban updates is high when the incentives are strong for urban updates (2004 to 2007) and
this share decreases when the incentives are strong for rural updates (after 2008). Figure 2
shows the percentage of properties up to date and confirms that the targets were binding
constraints for IGAC throughout the sample period.
I further use this variation in the supply of updates to look for evidence on selection into
cadastral updating. I consider the possibility that municipalities are intentionally updating
to collect more tax revenue and I try to get a sense of the size of this selection effect by
comparing the effect of updating on tax revenue across update cohorts. This exercise is
motivated by the large variation in number and type of updates shown in Figure 1, which
potentially reflects large differences in the composition of the update pool. The results, shown
in Figure 3, indicate that the returns to updating are fairly homogeneous across cohorts. I
am unable to reject the null hypothesis that the return is actually the same for all cohorts
between 2002 and 2011 at conventional levels of significance.
The final piece of evidence on the exogeneity of cadastral updates exploits IGAC’s yearly
planning of the updates. At the start of each year, IGAC drafts a list of municipalities to
be updated in that year, which the regional offices then use to negotiate with the municipal
and departmental governments regarding feasibility, funding and logistics. These lists are not
publicly available but I had access to the ones for 2011 and 2012. Matching these lists with the
actual updates that took place each year, I find that 80 % of updaters were in IGAC’s initial
list and 68 % of those pre-selected actually updated. Furthermore, results from regressions
using a specification similar to equation 1 reveal that the only robust predictors of inclusion
in the initial list are the number of years since the last urban and rural updates.
Taken together, the available evidence indicates that municipalities have a limited ability
to manipulate the timing of the cadastral updates and that this timing is mainly driven
by the supply from IGAC. However, municipalities have discretion over how much taxes to
collect. This autonomy over tax collection could be a problem if, for instance, only the mu-
nicipalities with good investment opportunities collect more taxes after a cadastral update.
Figure 3 already suggests that municipal governments do not enjoy large discretion over tax
collection. I provide evidence on statutory tax rates using data from Iregui et al. (2003) for
211 municipalities between 1999 and 2002. I regress the statutory property tax rate on a
dummy for the years after a cadastral update plus municipality and year fixed effects. The
15
results are shown in Table 4. The point estimates are very small and statistically insignifi-
cant, indicating that municipalities do not adjust statutory rates in response to cadastral
updates.26 Figure 4 further confirms that roughly 75 % of updaters between 2006 and 2010
experience an increase in property tax revenue following an update. The second part of the
identification strategy exploits plausibly exogenous variation in the world price of oil and the
heterogeneous distribution of this resource across municipalities. This type of difference-in-
differences methodology has been widely used in recent studies on Colombia.27 Identification
comes from the assumption that the world price of oil is exogenous to local conditions in
Colombian municipalities. This seems plausible, as Colombia is a relatively small exporter of
oil. According to the US Energy Information Administration, Colombia is the 18th largest
exporter of oil with less than 1 % of world exports.
I also assume that any systematic differences between municipalities with different levels
of oil abundance are time-invariant and thus captured by the municipality fixed effects.
As a measure of oil abundance I use the average amount of oil royalties received by the
municipality between 2000 and 2004 (royaltiesoili,00−04). Hence, what I exploit is the differential
impact of variation in the price of oil in municipalities with varying levels of oil extraction
in previous years (since royalties are a linear function of output). The variation from oil
discoveries, for example, is not exploited by this research design. Additionally, by using a
5-year average of previous oil royalties I address potential concerns about regression to the
mean.
Figure 7a summarizes the identification strategy for royalties. The black line shows the oil
price index I construct (2004=1). There is significant variation in oil prices during the sample
period: the price increases up to 2008, crashes in 2009 as a result of the global financial crisis
and recovers in the last two years. The figure also shows point estimates and 95 % confidence
intervals from a regression of the form
royaltiesi,j,t = αi + δj,t +2011∑
k=2006
βk [D(year = k)t ×D(oil royalties > 0)i,00−04] + εi,j,t (2)
where the dependent variable is royalties per capita in municipality i from department j
in year t. The coefficients of interest βk capture the average royalties among the set of
oil-rich municipalities (positive oil royalties between 2000 and 2004), net of municipality
and department-year fixed effects. The results in Figure 7a show that royalties in these
municipalities track the yearly variation in the price of oil.
26Sanchez and Espana (2013) provide additional evidence on the stickiness of statutory rates based oninterviews with public officials from several municipalities.
27See, for example, Dube and Vargas (2013); Carreri and Dube (2015); Santos (2014); Idrobo et al. (2014).
16
The map in Figure 5 shows sextiles of the distribution of (positive) average oil royalties
between 2000 and 2004. Two things stand out. First, although oil royalties are geographically
clustered in areas where there is oil, there is still substantial within-cluster variation in oil
intensity. The inclusion of department-year fixed effects ensures that I only exploit this
within-department variation in oil intensity. Second, a comparison with Figure 6 shows that
there is substantial overlap between oil-royalty recipients and municipalities with a cadastral
update. This allows me to verify that any differential effects across sources of revenue are
not driven by systematic differences in the use of revenue across municipalities, no matter
the source.
In what follows I use two main specifications. I show reduced-form effects of cadastral
updating and oil-price shocks using the following model:
yi,j,t = αi + δj,t + γTD(post-update)i,t + γR[priceoil
t × royaltiesoili,00−04
]+ εi,j,t (3)
where yi,j,t is an outcome of interest and αi and δj,t are municipality and department-year
fixed effects, respectively. The standard errors are clustered two-way by municipality and
department-year following Cameron et al. (2011). I standardize the “predicted” royalties
variable priceoilt × royaltiesoil
i,00−04 for ease of interpretation.
I estimate the effects of tax revenue and royalties on the outcomes of interest using an
instrumental variables model:
yi,j,t = αi + δj,t + βT property tax revenuei,t + βR natural resource royaltiesi,t + ui,j,t (4)
where the cadastral update dummy and the interaction of the oil price with oil intensity are
used as instruments for tax revenue and royalties, respectively.
To account for the fact that there may be a lag in the expenditure of royalties, I some-
times modify equations 3 and 4 and use cumulative royalties (∑t
k=2006 royaltiesi,k) instead
of contemporary values. The point here is that perhaps outcomes in year t are not so much
affected by royalties in that year but by the amount of royalties received in the previous
years. The municipality fixed effect absorbs all royalties up to 2005 and so these cumulatives
are partial ones since 2005. As an instrument for cumulative royalties I use the cumulative
of predicted royalties:∑t
k=2006 royaltiesoili,00−04 × priceoil
k .
Due to data constraints I change the unit of observation to municipality-local political
period for the regressions on disciplinary processes. I use data from the periods 2001-2003,
2004-2007 and 2008-2011. I use the oil royalties from 2000, supplied by Ecopetrol, as an
indicator of oil intensity.
17
5. Main Results
Table 5 shows the first stage results. Column 1 shows that cadastral updating leads on
average to a statistically significant increase of $6,000 pesos per capita in property tax re-
venue. Column 2 shows that a one standard deviation increase in predicted oil royalties due
to oil-price shocks leads to a $182,000 pesos increase in royalties, which is also statistically
significant. Although I find that both instruments are strong predictors of their respective
source of revenue, it is clear that cadastral updating has a much smaller effect on the pu-
blic purse than oil royalties. To put these numbers into perspective, total revenue averages
$500,000 pesos per capita, as shown in Table 2. I return to this later.
Table 6 shows the main results of the paper. Panel A shows reduced-form estimates
of the effect of the instruments on the outcomes of interest. The dependent variable is
specified in the header of each column and they are all in logs. The results indicate that a
cadastral update leads to a 0.8 % increase in educational enrolment and to a 12 % reduction
in the water contamination index. The effect on the percentage of poor population with
subsidized health insurance is also positive (1.2 % increase), but not statistically significant.
Surprisingly, additional tax revenue seems to lead to an increase in infant mortality, but the
effect is very small and also statistically insignificant. On the other hand, a one deviation
increase in predicted royalties leads to a 24 percent decrease in water contamination. There
is no evidence of any other effect.
Panel C shows the IV estimates. I find that a $1 million increase in property tax revenue
per capita leads to a 142 precent increase in educational enrolment and to a 14 percent
decrease in the water contamination index. These effects are larger than those of a $1 million
increase in royalties per capita by more than one order of magnitude and the difference is
statistically significant. The results for subsidized health insurance point in the same direction
but the difference is not statistically significant. Overall, there is no evidence that royalties
have a positive effect on any of the outcomes.
As mentioned above, the previous comparison could be affected by a lag in the effect
of royalties. I replicate the analysis using cumulative royalties in Table 7. The results on
tax revenue look the same while the results for royalties deteriorate significantly. The IV
estimates in panel B indicate that cumulative royalties lead to a worsening of all the outcomes
considered except infant mortality, with the estimates being statistically significant for access
to subsidized health insurance and water contamination. I present additional evidence from
yearly averages below that is consistent with these findings.
In Table 8, I show results from similar regressions using dummies for target achievement as
dependent variables. The reduced-form estimates in panel A indicate that a cadastral update
18
increases the probability of having universal coverage of poor population with subsidized
health insurance by 3 percentage points. This is a relatively large effect, given that only
15 % of municipalities met this target in 2005, and it is also statistically significant at the
10 % level. Column 4 additionally shows that a cadastral update leads to an increase of 7.6
percentage points in the probability that water in the municipality is suitable for human
consumption. When scaled by revenue using the IV estimator, I find that these effects on
health and water are significantly different from those of additional royalties. None of the
point estimates for royalties in panel B are statistically different from zero. Results using
cumulative royalties, shown in Table 9, lead to similar conclusions.
I next verify that the previous results are coming from different sources of revenue rat-
her than from different sets of municipalities, cadastral updaters and oil-royalty recipients.
In Table 10 I explore the possibility of heterogeneity in the effects of cadastral updating
depending on oil abundance, defined in the same way as in the estimations above: average
oil royalties between 2000 and 2004. The results show no evidence of heterogeneity in the
effects of cadastral updating by oil intensity. The point estimates for the interaction between
cadastral updating and oil abundance are very small and never statistically different from
zero. Furthermore, I can reject the null that the effect of updating is zero on educational
enrolment, water contamination and health insurance affiliation for the median and mean
levels of oil intensity.
The previous results provide systematic evidence that increases in local tax revenue
resulting from cadastral updates lead to larger improvements in local public good provision
than oil-price driven increases in royalties of the same magnitude. I provide evidence that
these results are driven by a differential response from local governments to increases in
revenue from these two sources using the data on disciplinary processes.
Table 11 shows the results. The reduced-form estimates in panel A indicate that higher
oil prices lead to a worsening of local government performance in municipalities that are
oil-royalty recipients. A one standard deviation increase in predicted royalties increases the
probability that the local mayor is prosecuted (25 percentage points), found guilty (22 per-
centage points) and removed from office (14 percentage points). Similar results are found for
top members of staff. Cadastral updating, on the other hand, has a negative effect on the
likelihood of these events, though the estimates are statistically insignificant.
The IV estimates point in the same direction: increases in royalties increase the probabi-
lity of disciplinary processes involving local public officials while increases in royalties reduce
this probability. The findings in columns 3 and 6 are very telling, as they indicate that the
offenses committed by local public officials are serious ones such as embezzlement and co-
rruption. These results provide evidence in support of changes in incumbent performance as
19
the underlying mechanism for the observed difference across sources of revenue in the effects
on development indicators.
6. Robustness and Extensions
One concern regarding the previous exercises is that royalties may be used for large-
scale projects with higher returns in the long run. Although the use of cumulative royalties
already addresses this concern, I provide additional evidence by running regressions similar to
equation 2, but using the outcomes of interest as dependent variables. The results are shown
in panels (c)-(f) of Figure 7. As already discussed, panel (a) shows that these oil-endowed
municipalities never receive less royalties than in 2005, and actually receive more between
2006 and 2008. Panel (b) then shows that they never spend less than in 2005, but they do
spend more in 2006 and 2007. However, panels (c)-(f) tell stories for the four outcomes of
interest that are consistent with the previous findings: despite the extra revenue and the
extra expenditure there is no observable improvement for any indicator. If anything, they
seem to worsen.
Another concern regarding the previous exercises is that changes in the price of oil may
affect the outcomes of interest in the municipalities where oil is extracted through other
channels besides government revenue. Since I am looking at the effect of higher oil prices in
municipalities that were already producing oil between 2000 and 2004, we should not expect
to observe the structural transformation associated with oil discoveries (Michaels, 2011).
Nevertheless, Table 12 shows that contemporary oil-price shocks are positively correlated
with FARC presence and, somewhat surprisingly, negatively correlated with the homicide
rate. Cumulative royalties, on the other hand, are positively correlated with population and
with business tax revenue, with this latter variable being a potential proxy for municipal
GDP (Sanchez and Nunez, 2000). These results are consistent with the idea that a series
of positive oil-price shocks lead to an economic boom in the municipality and this attracts
immigration.28
To address this concern I show that the results are unaffected by the inclusion of these
variables as controls. Figure 8 shows for each outcome of interest the point estimate and 95 %
confidence interval for royalties from equation 3, but also from a modified specification that
includes controls for ln population, business tax revenue, murder rate and FARC activity.
The figure shows that the estimates are remarkably robust to the inclusion of the control
variables. Although these are ‘bad controls’ in the sense of Angrist and Pischke (2009), the
28Several previous studies have exploited commodity-price shocks as a source of variation in local income(Miller and Urdinola, 2010; Dube and Vargas, 2013; Acemoglu et al., 2013; Asher and Novosad, 2014).
20
robustness of the estimates on royalties to their inclusion indicates that these variables are
not mediating the estimated effects.
I further explore the role of non-fiscal effects of oil-price shocks by exploiting the cross-
sectional variation in oil intensity. Figure 9 shows yearly average total revenue (panel A)
and royalties (panel B) for each quartile of the distribution of average 2000-2004 royalties,
as well as for municipalities that did not receive oil royalties in this period. The takeaway
from the graph is that municipalities in the top quartile are much richer than all other
municipalities and that this extra revenue is clearly coming from royalties. Municipalities in
the third quartile, on the other hand, appear to be much more comparable to the rest of the
country.
In Table 13 I replicate some of the previous analyses, but looking at the effect of oil price
shocks separately for municipalities in the third and fourth quartiles. Column 1 shows across
all panels that these shocks have a positive effect on royalties. However, columns 2-5 provide
evidence of heterogeneous non-fiscal effects across the two top quartiles. In particular, the
correlation with business tax revenue and FARC activity is only present for the top quartile
and the effect on population is much stronger for this group of municipalities. Overall, the
non-fiscal effects of oil-price shocks appear to be much weaker in municipalities belonging
to the third quartile. However, the results in columns 6-9 provide no robust evidence of a
reduced-form effect on the outcomes of interest in either quartile. The yearly averages for
each quartile shown in Figure 10 point in the same direction. These results (or the lack of)
provide further evidence on the robustness of the main results to potential violations of the
exclusion restriction for the oil instrument. They also suggest that the differential effects
of tax revenue and royalties on local public goods are not driven by the difference in scale
between the amounts of revenue from each source.
Another concern is that cadastral updates lead to a stable increase in tax revenue while
price shocks lead to unpredictable and potentially large variation in oil royalties. The results
on cumulative royalties, which don’t decrease, suggest that this difference is not behind the
results. Perhaps more important is the possibility that local governments are more prudent
in the way they spend these occasional resource windfalls. If they are more prudent, we
should observe that the propensity to spend the marginal peso of royalties is smaller than
the propensity to spend the marginal peso of taxes. Hence, I look at expenditure next.
In Table 14 I show results from estimating equation 4 using total expenditure per capita
as dependent variable and lagging royalties one year. Separate regressions for each source of
revenue in columns 1 and 2 reveal that one extra peso of tax revenue leads to approximately
1.3 extra pesos of expenditure, while one extra peso of royalties leads to 0.6 extra pesos of
expenditure. Although the point estimate for tax revenue is more than twice as large as that
21
for royalties, the standard errors are quite large and I fail to reject the null hypothesis that
either of the coefficients is different from one. The results are fairly similar if I look at both
sources in the same regression: even though the point estimate for tax revenue increases to
1.8, I still fail to reject the null that the coefficients are both equal to 1. Furthermore, column
4 shows that by using the cumulative royalties instead, which impose less structure on the
timing of expenditure than the lag, the coefficient for royalties rises to 1.2. This coefficient
is much closer to the estimates for tax revenue and is, once again, not statistically different
from 1. Overall, the available evidence indicates that one extra peso from either source leads
to one extra peso of expenditure.
Table 15 further explores the way revenue is spent. I replicate the analysis from Table 14
separately for operating expenses and investment, which are the two main broad expenditure
categories in the data. The estimates for investment in columns 1-4 resemble substantially
the results for total expenditure and I cannot reject the null hypothesis that the propensity
to spend out of both sources is the same and is equal to 1. The point estimates for operating
costs in columns 5-8 are much smaller and still statistically equivalent across sources. Since
the creation of jobs would lead to an increase in wages paid by the municipality, these results
suggest that revenue is not being used for patronage. Taken together, the evidence indicates
that extra revenue from either source is equally spent and it is spent almost exclusively on
investment. Investment here is defined as Gross Fixed Capital Formation, which means that
the money is being channeled for the most part towards infrastructure.
Why is investment of royalties not translating into improvements in the outcomes? The
results on disciplinary offences above already suggest that the quality of contracts is lower for
projects financed with royalties than for projects financed with taxes. Several of the removals
from office of mayors in oil-royalty recipient municipalities were related to violations of
procurement and public contracting laws. In Table 16 I provide further evidence on the lower
quality of investment with royalties by looking at educational infrastructure. The results in
column 1 show that an increase in property tax revenue leads to an increase in the number
of schools in the municipality, while an increase in royalties has no effect. Consistently with
this result, the point estimates for teaching area are much larger for tax revenue in column
3 than for royalties, although the difference is not statistically significant.
7. Conclusion
This paper sheds light on whether fiscal decentralization can succeed in a context of
high dependence on external sources of revenue. I show that locally-raised tax revenue has
a larger impact on the provision of public goods than royalties from the extraction of oil
22
in Colombian municipalities. There is, in fact, no robust evidence of an effect of royalties
on public services over the period 2005-2011, despite the fact that royalties are earmarked
for this purpose. I complement these findings with evidence from disciplinary processes that
indicates that internal and external revenue have opposite effects on the misbehavior of local
public officials.
Where does the power of taxation lie? I am unable to distinguish between an explanation
based on information, where voters are better informed about tax revenue, and one based
on preferences, where voters dislike taxation. Future research must try to better understand
the relative importance of these two mechanisms, as this could determine what the optimal
policy to address the limitations of current decentralization schemes is. The experimental
work in Paler (2013) and Martin (2014) constitutes early steps in this direction, although it
would be desirable to obtain evidence from outside the lab on this matter. Another avenue
for future research is related to the study of different tax instruments with the objective of
establishing which characteristics (e.g. salience) are particularly important for accountability.
Nevertheless, these considerations are of second order relative to the fact that the return
to external revenue is much lower than that of tax revenue. The findings of this paper have
implications for important policy debates regarding fiscal federalism and the management of
natural resource wealth. They are also relevant for discussions on foreign aid. Overall, these
findings invite policymakers to reconsider the benefits from the disbursement of resources to
local governments without a locally-financed counterpart. They also suggest that investments
in local fiscal capacity may have much higher returns than previously expected.
References
Acemoglu, Daron, Amy Finkelstein, and Matthew J. Notowidigdo (2013), “Income and
health spending: Evidence from oil price shocks.” The Review of Economics and Statistics,
95 (4), 1079–1095.
Alesina, Alberto, Alberto Carrasquilla, and Juan Jose Echevarria (2005), “Decentralization
in Colombia.” In Institutional reforms in Colombia (Alberto Alesina, ed.), MIT Press,
Cambridge, MA.
Alesina, Alberto and Guido Tabellini (2007), “Bureaucrats or politicians? part I: A single
policy task.” American Economic Review, 97 (1), 169–179.
Angrist, Joshua D. and Jorn-Steffen Pischke (2009), Mostly Harmless Econometrics: an em-
piricist’s companion. Princeton University Press, Princeton, NJ.
23
Asher, Samuel and Paul Novosad (2014), “Digging for development: Mining booms and local
economic development in India.” Working Paper.
Banerjee, Abhijit, Selvan Kumar, Rohini Pande, and Felix Su (2011), “Do informed voters
make better choices? experimental evidence from urban India.” Working Paper.
Bardhan, Pranab (2002), “Decentralization of governance and development.” Journal of
Economic Perspectives, 16 (4), 185–205.
Bardhan, Pranab and Dilip Mookherjee (2006), “Decentralization, corruption and govern-
ment accountability: An overview.” In International Handbook on the economics of co-
rruption (Susan Rose-Ackerman, ed.), Edward Elgar.
Bauer, Peter (1972), Dissent on development. Harvard University Press, Cambridge (Mass.).
Beblawi, Hazem (1990), “The rentier state in the arab world.” In The Arab State (Giacomo
Luciani, ed.), University of California Press, Berkeley.
Besley, Timothy and Robin Burgess (2002), “The political economy of government respon-
siveness: Theory and evidence from India.” The Quarterly Journal of Economics, 117 (4),
1415–1451.
Besley, Timothy and Maitreesh Ghatak (2006), “Public goods and economic development.”
In Understanding Poverty (Abhijit Banerjee, Roland Benabou, and Dilip Mookherjee,
eds.), Oxford University Press.
Besley, Timothy and Torsten Persson (2011), Pillars of Prosperity: The Political Economics
of Development Clusters. Princeton University Press, Princeton, New Jersey.
Besley, Timothy and Torsten Persson (2013), “Taxation and development.” In Handbook of
public economics (Martin Feldstein Alan J. Auerbach, Raj Chetty and Emmanuel Saez,
eds.), volume 5, 51–110, Elsevier.
Besley, Timothy and Torsten Persson (2014), “Why do developing countries tax so little?”
Journal of Economic Perspectives, 28 (4), 99–120.
Bjorkman, Martina and Jakob Svensson (2009), “Power to the people: Evidence from a
randomized field experiment on community-based monitoring in Uganda.” The Quarterly
Journal of Economics, 124 (2), 735–769.
Borge, Lars-Erik, Pernille Parmer, and Ragnar Torvik (2015), “Local natural resource curse?”
Forthcoming, Journal of Public Economics.
24
Borge, Lars-Erik and Jørn Rattsø (2008), “Property taxation as incentive for cost control:
Empirical evidence for utility services in Norway.” European Economic Review, 52 (6),
1035–1054.
Brollo, Fernanda, Tommaso Nannicini, Roberto Perotti, and Guido Tabellini (2013), “The
political resource curse.” American Economic Review, 103 (5), 1759–1796.
Brutti, Zelda (2015), “Cities drifting apart: heterogeneous outcomes of decentralizing public
education.” Working Paper.
Cabral, Marika and Caroline Hoxby (2015), “The hated property tax: salience, tax rates and
tax revolts.” NBER Working Paper 18514.
Cameron, Colin, Jonah Gelbach, and Douglas Miller (2011), “Robust inference with multi-
way clustering.” Journal of Business and Economic Statistics, 29, 238–249.
Carreri, Maria and Oeindrila Dube (2015), “Do natural resources influence who comes to
power, and how?” Working Paper.
Casaburi, Lorenzo and Ugo Troiano (2015), “Ghost-house busters: The electoral response to
a large anti tax evasion program.” The Quarterly Journal of Economics, Forthcoming.
Caselli, Francesco (2006), “Power struggles and the natural resource curse.” Unpublished.
Caselli, Francesco and Tom Cunningham (2009), “Leader behaviour and the natural resource
curse.” Oxford Economic Papers, 61 (4), 628–650.
Caselli, Francesco and Guy Michaels (2013), “Do oil windfalls improve living standards?
evidence from Brazil.” American Economic Journal: Applied Economics, 5 (1), 208–38.
Chetty, Raj, Adam Looney, and Kory Kroft (2009), “Salience and taxation: Theory and
evidence.” American Economic Review, 99 (4), 1145–77.
Chong, Alberto, Ana De la O, Dean Karlan, and Leonard Wantchekon (2015), “Does co-
rruption information inspire the fight or quash the hope? a field experiment in Mexico on
voter turnout, choice and party identification.” The Journal of Politics, 77 (1), 55–71.
Collier, Paul (2006), “Is aid oil? an analysis of whether Africa can absorb more aid.” World
Development, 34 (9), 1482 – 1497.
Cortes, Darwin (2010), “Do more decentralized local governments do better? an evaluation
of the 2001 decentralization reform in Colombia.” Universidad del Rosario, Documento de
Trabajo 84.
25
De Janvry, Alain, Frederico Finan, and Elisabeth Sadoulet (2012), “Local electoral incentives
and decentralized program performance.” The Review of Economics and Statistics, 94 (3),
672–685.
De la O, Ana (2013), “Do conditional cash transfers affect electoral behavior? evidence from
a randomized experiment in Mexico.” American Journal of Political Science, 57 (1), 1–14.
Dewatripont, Mathias, Ian Jewitt, and Jean Tirole (1999), “The economics of career concerns,
part I: comparing information structures.” Review of Economic Studies, 66 (1), 183–198.
Dollery, Brian E. and Andrew C. Worthington (1996), “The empirical analysis of fiscal
illusion.” Journal of Economic Surveys, 10 (3), 261–297.
Dube, Oeindrila and Juan F. Vargas (2013), “Commodity price shocks and civil conflict:
Evidence from Colombia.” Review of Economic Studies, 80 (4), 1384–1421.
Easterly, William (2006), The white man’s burden: Why the West’s efforts to aid the rest
have done so much ill and so little good. Penguin, New York City.
Faguet, Jean-Paul (2014), “Decentralization and governance.” World Development, 53, 2–13.
Faguet, Jean-Paul and Fabio Sanchez (2008), “Decentralization’s effects on educational out-
comes in Bolivia and Colombia.” World Development, 36 (7), 1294–1316.
Faguet, Jean-Paul and Fabio Sanchez (2014), “Decentralization and access to social services
in Colombia.” Public Choice, 160 (1-2), 227–249.
Fehr, Ernst and Simon Gachter (2000), “Fairness and retaliation: The economics of recipro-
city.” Journal of Economic Perspectives, 14 (3), 159–181.
Fergusson, Leopoldo, Juan F. Vargas, and Mauricio Vela (2013), “Sunlight disinfects? free
media in weak democracies.” Documento CEDE 2013-14.
Ferraz, Claudio and Frederico Finan (2008), “Exposing corrupt politicians: The effects of
Brazil’s publicly released audits on electoral outcomes.” The Quarterly Journal of Econo-
mics, 123 (2), 703–745.
Ferraz, Claudio and Frederico Finan (2011), “Electoral accountability and corruption: Evi-
dence from the audits of local governments.” American Economic Review, 101 (4), 1274–
1311.
Ferraz, Claudio and Joana Monteiro (2014), “Learning to punish: Resource windfalls and
political accountability in Brazil.” Working Paper.
26
Finkelstein, Amy (2009), “E-ztax: Tax salience and tax rates.” The Quarterly Journal of
Economics, 124 (3), 969–1010.
Fisman, Raymond and Roberta Gatti (2002), “Decentralization and corruption: Evidence
from U.S. federal transfer programs.” Public Choice, 113 (1-2), 25–35.
Fujiwara, Thomas (2015), “Voting technology, political responsiveness, and infant health:
Evidence from Brazil.” Econometrica, 83 (2), 423–464.
Gadenne, Lucie (2015), “Tax me but spend wisely: Sources of public finance and government
accountability.” Working Paper.
Gadenne, Lucie and Monica Singhal (2014), “Decentralization in developing economies.”
Annual Review of Economics, 6 (1), 581–604.
Glaeser, Edward L. (1996), “The incentive effects of property taxes on local governments.”
Public Choice, 89 (1/2), 93–111.
Glaeser, Edward L. (2013), “Urban public finance.” In Handbook of public economics (Mar-
tin Feldstein Alan J. Auerbach, Raj Chetty and Emmanuel Saez, eds.), volume 5, 195–256,
Elsevier.
Guiteras, Raymond and Ahmed Mushfiq Mobarak (2014), “Does development aid undermine
political accountability? leader and constituent responses to a large-scale intervention.”
Working Paper.
Herrera, Juan Camilo (2014), “Petroleo y desarrollo: Efectos en la acumulacion y destruccion
de capital humano de los municipios de Colombia.” Documento CEDE, 2014 (10).
Holmstrom, Bengt (1999), “Managerial incentive problems: A dynamic perspective.” Review
of Economic Studies, 66 (1), 169–182.
Hoxby, Caroline (1999), “The productivity of schools and other local public goods produ-
cers.” Journal of Public Economics, 74 (1), 1–30.
Idrobo, Nicolas, Daniel Mejıa, and Ana Marıa Tribın (2014), “Illegal gold mining and violence
in Colombia.” Peace Economics, 20 (1), 83–111.
Iregui, Ana Marıa, Ligia Melo, and Jorge Ramos (2003), “El impuesto predial en Colombia:
evolucion reciente, comportamiento de las tarifas y potencial de recaudo.” Borradores de
Economıa 274, Banco de la Republica.
27
Iregui, Ana Marıa, Ligia Melo, and Jorge Ramos (2004), “El impuesto predial en Colombia:
Factores explicativos del recaudo.” Borradores de Economıa 319, Banco de la Republica.
Leigh, Andrew (2009), “Does the world economy swing national elections?” Oxford Bulletin
of Economics and Statistics, 71 (2), 163–181.
Litschig, Stephan and Kevin M. Morrison (2013), “The impact of intergovernmental trans-
fers on education outcomes and poverty reduction.” American Economic Journal: Applied
Economics, 5 (4), 206–40.
Mahdavy, Hossein (1970), “The patterns and problems of economic development in rentier
states: The case of Iran.” In Studies in the Economic History of the Middle East from the
Rise of Islam to the present day (M.A. Cook, ed.), Oxford University Press, London.
Maldonado, Stanislao (2014), “The political effects of resource booms: Political outcomes,
clientelism and public goods provision in Peru.” Working Paper.
Martin, Lucy (2014), “Taxation, loss aversion and accountability: Theory and experimental
evidence for taxation’s effect on citizen behavior.” Working Paper.
Martınez-Bravo, Monica, Gerard Padro i Miquel, Nancy Qian, and Yang Yao (2014), “Poli-
tical reform in China: the effect of local elections.” Working Paper.
Matsen, Egil, Gisle Natvik, and Ragnar Torvik (2015), “Petro Populism.” Forthcoming,
Journal of Development Economics.
Mehlum, Halvor, Karl Moene, and Ragnar Torvik (2006), “Institutions and the resource
curse.” The Economic Journal, 116 (508), 1–20.
Michaels, Guy (2011), “The long term consequences of resource-based specialisation.” The
Economic Journal, 121 (551), 31–57.
Miller, Grant and B. Piedad Urdinola (2010), “Cyclicality, mortality, and the value of time:
The case of coffee price fluctuations and child survival in Colombia.” Journal of Political
Economy, 118 (1), 113–155.
Mookherjee, Dilip (2015), “Political decentralization.” Annual Review of Economics, 7 (1),
231–249.
North, Douglass C. and Barry R. Weingast (1989), “Constitutions and commitment: The
evolution of institutional governing public choice in seventeenth-century England.” The
Journal of Economic History, 49 (4), 803–832.
28
Oates, Wallace (2005), “Toward a second-generation theory of fiscal federalism.” Internatio-
nal Tax and Public Finance, 12, 349–373.
Olsson, Ola and Michele Valsecchi (2014), “Resource windfalls and public goods: Evidence
from a policy reform in indonesia.” Working Paper.
Paler, Laura (2013), “Keeping the public purse: An experiment in windfalls, taxes, and the
incentives to restrain government.” American Political Science Review, 107 (4), 706–725.
Persson, Torsten and Guido Tabellini (2000), Political Economics: Explaining Economic
Policy. MIT Press, Cambridge (Mass.).
Pritchett, Lant and Yamini Aiyar (2014), “Value subtraction in public sector production:
Accounting versus economic cost of primary schooling in India.” Center for Global Deve-
lopment Working Paper, 2014 (391).
Reinikka, Ritva and Jakob Svensson (2004), “Local capture: Evidence from a central go-
vernment transfer program in Uganda.” The Quarterly Journal of Economics, 119 (2),
679–705.
Reinikka, Ritva and Jakob Svensson (2005), “Fighting corruption to improve schooling: Evi-
dence from a newspaper campaign in Uganda.” Journal of the European Economic Asso-
ciation, 3 (2-3), 259–267.
Robinson, James A., Ragnar Torvik, and Thierry Verdier (2006), “Political foundations of
the resource curse.” Journal of Development Economics, 79 (2), 447–468.
Ross, Michael L. (2001), “Does oil hinder democracy?” World Politics, 53, 325–361.
Ross, Michael L. (2004), “Does taxation lead to representation?” British Journal of Political
Science, 34 (2), 229–249.
Santos, Rafael (2014), “Not all that glitters is gold: Gold boom, child labor and schooling in
colombia.” Documento CEDE, 2014 (31).
Snyder, James M. and David Stromberg (2010), “Press coverage and political accountability.”
Journal of Political Economy, 118 (2), 355–408.
Sanchez, Fabio and Irina Espana (2013), “Estructura, potencial y desafıos del impuesto
predial en Colombia.” Documento CEDE, 2013 (48).
Sanchez, Fabio and Jairo Nunez (2000), “La geografıa y el desarrollo economico: Una apro-
ximacion municipal.” Desarrollo y Sociedad, 46, 43–108.
29
Sanchez, Fabio and Monica Pachon (2013), “Descentralizacion, esfuerzo fiscal y progreso
social en Colombia en el nivel local, 1994-2009: ¿por que importa la polıtica nacional?”
Documento CEDE, 2013 (38).
Sanchez, Fabio and Alexander Vega (2014), “Cobertura de acueducto y alcantarillado, ca-
lidad del agua y mortalidad infantil en Colombia, 2000-2012.” Documento CEDE, 2014
(40).
Tribın, Ana Marıa (2014), “Chasing votes with the public budget.” Working Paper.
Vicente, Pedro C. (2010), “Does oil corrupt? evidence from a natural experiment in West
Africa.” Journal of Development Economics, 92 (1), 28–38.
Weingast, Barry R. (2009), “Second generation fiscal federalism: The implications of fiscal
incentives.” Journal of Urban Economics, 65 (3), 279–293.
World Bank (2000), World Development Report 1999/2000: Entering the 21st century. Ox-
ford University Press.
World Bank (2004), World Development Report 2004: Making services work for poor people.
The World Bank and Oxford University Press.
Zhuravskaya, Ekaterina V. (2000), “Incentives to provide local public goods: fiscal federalism,
russian style.” Journal of Public Economics, 76 (3), 337–368.
30
Figure 1: Cadastral updates per year and presidential term
Note: I assign to each administration its first full calendar year (since presidential terms always start on August 7th) and thethree following ones. Each term is then pushed back by one year to account for the fact that updated cadastres only come intoeffect the 1st of January of the following year.
Figure 2: Percentage of properties with up-to-date cadastres
31
Figure 3: Property tax revenue increases after a cadastral update by cohort
The graph shows point estimates and 95 % confidence intervals for a regression of ln property tax revenue per capita (2004COP) on a set of dummies equal to one from year t onward if the municipality has a cadastral update on year t, weighted bythe share of the cadastre that was updated (depending on whether the update was urban, rural or both). I use property valuesfrom 2000 to determine the shares. The regression includes municipality and department-year fixed effects. Standard errorsclustered two-way by municipality and department-year.
Figure 4: Change in property tax revenue after a cadastral update
The graph shows the histogram for the difference between the average property tax revenue collected after a cadastral updateand the amount for the year before the update. Top and bottom 1 % removed for ease of visualization. Property tax revenue in2004 COP per capita. Update cohorts from 2006 to 2010. Pre-year corresponds to the first one in the case of multiple updates.
32
Figure 5: Municipalities receiving oil royalties between 2000 and 2004
Note: The map shows average oil royalties between 2000-2004. The lighter shade corresponds to municipalities that did notreceive oil royalties in this period while the darker shades correspond to sixtiles of the average oil-royalty distribution amongroyalty recipients. Areas in grey are excluded from the sample. This includes municipalities with their own cadastral agencies(Bogota, Medellin, Cali, Antioquia) and non-municipalized territories.
33
Figure 6: Municipalities with cadastral update (2006-2010)
Note: The map shows the location of municipalities that updated some portion of their cadastre between 2006 and 2010.Areasin grey are excluded from the sample. This includes municipalities with their own cadastral agencies (Bogota, Medellin, Cali,Antioquia) and non-municipalized territories.
34
Figure 7: Medium-run impact of oil price shocks
(a) Royalties (b) Expenditure
(c) Educational enrolment (d) Infant mortality
(e) Health insurance (f) Water contamination
Note: Each graph shows point estimates and 95 % confidence intervals from a regression of the variablein the caption on a set of year interactions (2006-2011) with a dummy for municipalities with positive oilroyalties between 2000 and 2004. These regressions use data from the period 2005-2011. All regressions includemunicipality and department-year fixed effects. Standard errors are clustered two-way by municipality anddepartment-year. In panel (a), the dark line shows the oil price index (2004=1), which I construct usingthe average petroleum spot price (IMF/IFS), the official exchange rate from Banco de la Republica and theConsumer Price Index calculated by DANE.
35
Figure 8: Impact of royalties with and without controls
(a) Educational enrolment (t) (b) Educational enrolment (cum.)
(c) Infant mortality (t) (d) Infant mortality (cum.)
(e) Health insurance (t) (f) Health insurance (cum.)
(g) Water containation (t) (h) Water contamination (cum.)
Note: Each graph shows point estimates and 95 % confidence intervals from a regression of the variable in thecaption on royalties, instrumented using oil price shocks. All regressions use data from the period 2005-2010.Panels on the left show results for contemporary royalties, while those on the right show the results forcumulative royalties. All regressions include municipality and department-year fixed effects. Standard errorsare clustered two-way by municipality and department-year.
36
Figure 9: Total revenue and royalties by average 2000-2004 oil royalties
Note: The figures show average total revenue (panel A) and avearge royalties (panel B) for each quartileof the average 2000-2004 positive oil royalties distribution. It also shows this information for municipalitieswith no oil royalties between 2000 and 2004. All money values in millions of 2004 COP per capita.
37
Figure 10: Medium-run impact of oil price shocks for the top two quartiles of oil abundance
(a) Royalties (b) Expenditure
(c) Educational enrolment (d) Infant mortality
(e) Health insurance (f) Water contamination
Note: Each graph shows point estimates and 95 % confidence intervals from a regression of the variablein the caption on a set of year interactions (2006-2011) with a dummy for municipalities with positive oilroyalties between 2000 and 2004. These regressions use data from the period 2005-2011. All regressions includemunicipality and department-year fixed effects. Standard errors are clustered two-way by municipality anddepartment-year. In panel (a), the dark line shows the oil price index (2004=1), which I construct usingthe average petroleum spot price (IMF/IFS), the official exchange rate from Banco de la Republica and theConsumer Price Index calculated by DANE.
38
Table 1: Achievement of targets by oil-royalty recipients
(1) (2) (3)Target1 Mean Target met
Indicator (2005) ( % in 2005)Net enrollment rate in basic education ( %) 100 91.8 30.0Infant mortality rate (h) 16.5 26.8 7.1Poor population with subsidized health insurance ( %) 100 72.7 14.3Contamination of drinking water index (0-100)2 5 32.2 17.9Population with access to drinking water ( %)3 70 63.0 62.1Population with access to sewerage ( %)3 70 41.1 26.4Note: The table shows the indicators on which at least 75 % of royalties have to be spent and the tar-gets that royalty recipients must meet. It also shows the average of each indicator in 2005 for the 140municipalities with positive oil royalties between 2000 and 2004, as well as the percentage of these muni-cipalities meeting the target. 1 Targets from Decree 1747/1995, modified by Law 1151/2007, Resolution4911/2008 (Education) and Decree 1447/2010 (Water). 2 Information from 2007, which is the first yearfor which data on the IRCA water contamination index is available. 3 Data on access to drinking waterand sewerage is only available from the 2005 population census.
39
Table 2: Summary statistics
Variable Mean Std. Dev. Min. Max. NA. population and cadastral updating
Population (thousands) 30.03 75.55 0.88 1193.67 6704Rural share of population 0.58 0.24 0 0.98 6704Cadastral update (dummy) 0.13 0.34 0 1 6704Cadastral valuation 3.59 4.19 0 83.16 6704Number of properties (thousands) 9.10 20.26 0 304.51 6704
B. oil price and royaltiesOil royalties 2000-2004 0.02 0.16 0 3.48 969Oil price (millions of 2004 COP per barrel) 0.13 0.02 0.1 0.15 7
C. public financeTotal revenue 0.54 0.43 0.09 6.38 6704Current revenue 0.14 0.12 0.02 2.13 6704Tax revenue 0.07 0.09 0 1.7 6704Property tax revenue 0.02 0.03 0 0.61 6704Capital revenue 0.4 0.36 0.03 5.68 6704Natural resource royalties 0.05 0.26 0 5.07 6704Transfers 0.36 0.23 0.05 5.53 6704Total expenditure 0.58 0.5 0.01 9.72 6704Current expenditure 0.08 0.11 0 7.76 6704Investment 0.5 0.45 0.01 9.31 6704
D. development indicatorsNet enrolment rate in basic education ( %) 88 17.18 18.7 244.4 6704Infant mortality rate (h) 22.81 8.51 9.24 64.09 6704Poor population with subsidized health insurance ( %) 87.15 15.66 0 100 6704IRCA water contamination index (0-100) 29.38 23.82 0 100 4472
E. disciplinary processesMayor prosecuted (dummy) 0.19 0.39 0 1 2985Mayor found guilty (dummy) 0.14 0.34 0 1 2985Mayor banned from office (dummy) 0.08 0.28 0 1 2985Top staff prosecuted (dummy) 0.06 0.24 0 1 2985Top staff found guilty (dummy) 0.04 0.2 0 1 2985Top staff banned from office (dummy) 0.03 0.17 0 1 2985Council member prosecuted (dummy) 0.05 0.22 0 1 2985Council member found guilty (dummy) 0.04 0.2 0 1 2985Council member banned from office (dummy) 0.03 0.17 0 1 2985
F. political characteristicsCandidates in last mayor election 3.81 1.85 1 17 2590Vote share for winning mayor 0.51 0.12 0.2 1 2590HHI last mayor election (votes) 0.26 0.12 0.03 1 2590Left-wing mayor (dummy) 0.03 0.18 0 1 2590Liberal mayor (dummy) 0.25 0.44 0 1 2590Conservative mayor (dummy) 0.22 0.42 0 1 2590Change of mayor’s party (dummy) 0.71 0.45 0 1 2375Parties/seat in last council election 0.56 0.22 0.08 1.57 2873HHI last council election (seats) 0.33 0.18 0.06 1 2873Share of council from mayor’s party 0.37 0.25 0 1 2582Share of council from left-wing parties 0.04 0.11 0 1 2873Vote share for winning governor 0.5 0.21 0 1 2917Mayor and governor from same party (dummy) 0.19 0.39 0 1 2382Vote share for winning president 0.54 0.21 0 1 2957HHI last House election (votes) 0.29 0.15 0 1 2965HHI last Senate election (votes) 0.26 0.14 0 0.94 2966
G. central-government policies, crime and conflictNew families enrolled in CCT program (per 10,000 inh.) 13.93 25.41 0 159.99 6704Uribe visit (dummy) 0.02 0.15 0 1 6704Municipality certified in education (dummy) 0.05 0.21 0 1 6704Value of new agricultural loans 85.3 95.84 0 1347.44 6704Murder rate (per 100,000 inh.) 33.87 42.87 0 532.75 6704Internal armed conflict (dummy) 0.26 0.44 0 1 6704FARC events (per 10,000 inh.) 1.37 4.88 0 98.7 5743ELN events (per 10,000 inh.) 0.23 1.47 0 32.62 5743AUC events (per 10,000 inh.) 0.07 0.61 0 25.65 5743Coca crops (dummy) 0.13 0.33 0 1 4797Note: The sample includes 969 municipalities for the period 2005-2011. Political characteristics in panel F are calculatedusing results from local elections in 2000, 2003 and 2007 and from national elections in 2002, 2006 and 2010. The variablesrelated to disciplinary processes in panel E contain information from the local political periods 2001-2003, 2004-2007 and2008-2011. All money variables are expressed in millions of 2004 Colombian pesos per capita, unless specified otherwise.
40
Table 3: Predicting a cadastral update
No controls Controls for recent updates
Variable Coefficient Standard error Coefficient Standard error Municipalities Observations(1) (2) (3) (4) (5) (6)
ln population 0.279 (0.178) 0.0190 (0.195) 969 6,704rural share of population -0.764 (0.707) -0.156 (0.617) 969 6,704ln non-property tax revenue 0.00215 (0.00990) 0.0105 (0.00975) 969 6,704ln total transfers -0.0135 (0.0235) 0.0143 (0.0225) 969 6,704natural resource royalties 3.79e-06 (2.70e-05) -2.00e-06 (3.36e-05) 969 6,704ln capital revenue -0.0148 (0.0173) -0.00397 (0.0166) 969 6,704# candidates last mayor election 0.00407 (0.00413) -0.00143 (0.00450) 968 6,160vote share for winning mayor 0.0284 (0.0588) 0.0732 (0.0558) 968 6,160HHI last mayor election (votes) 0.0231 (0.0608) 0.0436 (0.0582) 968 6,160D(left-wing mayor) 0.0270 (0.0313) 0.0146 (0.0331) 968 6,160D(Liberal mayor) -0.00530 (0.0185) 0.00487 (0.0182) 968 6,160D(Conservative mayor) 0.0220 (0.0179) 0.0272 (0.0167) 968 6,160D(change of mayor’s party) -0.00149 (0.0200) -0.000459 (0.0195) 790 5,432parties/seat last council election -0.0644* (0.0391) -0.0845** (0.0367) 969 6,624HHI last council election (seats) -0.0638 (0.0648) -0.0145 (0.0715) 969 6,624share of council from mayor’s party -0.0548 (0.0375) -0.0474 (0.0370) 965 6,136share of council from left-wing parties 0.0148 (0.0710) 0.0270 (0.0731) 969 6,624vote share for winning governor 0.0208 (0.0453) 0.0102 (0.0438) 969 6,678D(mayor and governor from same party) -0.0153 (0.0198) 0.00640 (0.0208) 968 6,154vote share for winning president 0.00457 (0.0588) 0.0152 (0.0585) 969 6,670HHI last House election (votes) 0.0549 (0.0617) 0.0229 (0.0586) 968 6,678HHI last Senate election (votes) 0.0157 (0.0493) 0.00896 (0.0462) 968 6,678D(Uribe visit) 0.0387 (0.0417) 0.0354 (0.0357) 969 6,704new families in CCT program 0.000191 (0.000276) 0.000147 (0.000254) 969 6,704value of new agricultural loans 2.72e-05 (8.88e-05) 6.73e-05 (9.40e-05) 969 6,704D(municipality certified in education) 0.0320 (0.0750) 0.0370 (0.0823) 969 6,704homicide rate -0.000120 (0.000159) -9.62e-05 (0.000160) 969 6,704FARC events -0.000452 (0.00168) 0.000264 (0.00151) 966 5,743ELN events -0.00375 (0.00324) -0.00486 (0.00351) 966 5,743D(coca crops) -0.00913 (0.0316) 0.0192 (0.0309) 969 4,797Each row corresponds to a different regression of a dummy equal to one the year before a cadastral update comes into effect onthe respective variable. All regressions include municipality fixed effects and department-year fixed effects. Columns 3 and 4 showestimates from an enlarged specification that also includes as controls separate dummy variables for the first five years after an urbanor rural update. Standard errors clustered two-way by municipality and by department-year. The number of observations varies dueto data availability. *** p<0.01, ** p<0.05, * p<0.1
Table 4: Cadastral updates and statutory tax rates
(1) (2) (3) (4)VARIABLES Property tax rate ln Property tax rate
D(post-cadastral-update)i,t 0.0964 0.176 0.0103 0.0185[0.151] [0.160] [0.0201] [0.0203]
Time fixed effects year dpt-year year dpt-yearObservations 799 799 799 799Number of municipalities 211 211 211 211Dependent variable in the header. All regressions include municipality fixed ef-fects. Columns 1 and 3 include year fixed effects while columns 2 and 4 includedepartment-specific year fixed effects. Sample period: 1999-2002. Average tax rateis 8.4 h. Standard errors clustered two-way by municipality and by department-year. *** p<0.01, ** p<0.05, * p<0.1
41
Table 5: First stage results
(1) (2) (3)Property Royalties Royalties
VARIABLES Tax (cumulative)
D(post-cadastral-update)i,t 0.00629***[0.00126]
royaltiesoili,00−04 × priceoil
t 0.182***[0.0393]∑t
k=2006 royaltiesoili,00−04 × priceoil
k 1.270***[0.0921]
Dependent variable mean 0.021 0.054 0.342- if oil-royalty recipient (00-04) - 0.249 1.668Observations 6,704 6,704 6,704Number of municipalities 969 969 969Dependent variable in the header. Money variables in millions of 2004COP per capita. All regressions include municipality and department-year fixed effects. Standardized coefficients are shown in columns 2 and3. Sample period: 2005-2011. Standard errors clustered two-way by mu-nicipality and department-year. *** p<0.01, ** p<0.05, * p<0.1
42
Table 6: Sources of revenue and public good provision
(1) (2) (3) (4)Educational Infant Health Water
VARIABLES enrolment mortality insurance contamination
PANEL A: REDUCED FORM
D(post-cadastral-update)i,t 0.00886*** 0.00295 0.0120 -0.123**[0.00334] [0.00210] [0.00850] [0.0560]
royaltiesoili,00−04 × priceoil
t 0.00254 0.000401 -0.00898 -0.239*[0.00217] [0.000485] [0.0161] [0.134]
PANEL B: IV
property tax revenuei,t 1.422** 0.472 1.894 -14.53*[0.616] [0.357] [1.363] [7.776]
natural resource royaltiesi,t 0.0231 0.00523 -0.0372 -1.133[0.0162] [0.00393] [0.0851] [0.805]
1st stage F-statistic 14.604 14.604 14.604 8.126p-value H0:tax=royalties 0.022 0.187 0.156 0.086
Dependent variable mean in 2005 (level) 86.1 24.2 74.3 30.5Observations 6,704 6,704 6,704 4,467Number of municipalities 969 969 969 937Dependent variable (natural log) in the header. Money variables in millions of 2004 COP per capita. Standardizedcoefficients for royaltiesoili,00−04 × priceoilt . In panel B, D(post-cadastral-update) and royaltiesoili,00−04 × priceoilt
are used as instruments for property tax revenue and natural resource royalties, respectively. All regressionsinclude municipality and department-year fixed effects (sample period: 2005-2011, except column 4: 2007-2011).Standard errors clustered two-way by municipality and department-year. *** p<0.01, ** p<0.05, * p<0.1
43
Table 7: Sources of revenue and public good provision (cumulative royalties)
(1) (2) (3) (4)Educational Infant Health Water
VARIABLES enrolment mortality insurance contamination
PANEL A: REDUCED FORM
D(post-cadastral-update)i,t 0.00886*** 0.00302 0.0127 -0.121**[0.00333] [0.00210] [0.00856] [0.0559]∑t
k=2006 royaltiesoili,00−04 × priceoil
k -0.000827 -0.00234 -0.0176** 0.131[0.00290] [0.00210] [0.00771] [0.0914]
PANEL B: IV
property tax revenuei,t 1.418** 0.483 2.027 -14.48*[0.619] [0.359] [1.378] [7.718]
natural resource royalties (cum.)i,t -0.00206 -0.00233 -0.0159*** 0.130**[0.00239] [0.00165] [0.00612] [0.0659]
1st stage F-statistic 11.807 11.807 11.807 7.381p-value H0:tax=royalties 0.022 0.176 0.139 0.059
Dependent variable mean in 2005 (level) 86.1 24.2 74.3 30.5Observations 6,704 6,704 6,704 4,467Number of municipalities 969 969 969 937Dependent variable (natural log) in the header. Money variables in millions of 2004 COP per capita. Standardizedcoefficients for
∑royaltiesoili,00−04×priceoilt . In panel B, D(post-cadastral-update) and
∑royaltiesoili,00−04×priceoilt
are used as instruments for property tax revenue and cumulative natural resource royalties, respectively. Allregressions include municipality and department-year fixed effects (sample period: 2005-2011, except column 4:2007-2011). Standard errors clustered two-way by municipality and department-year. *** p<0.01, ** p<0.05, *p<0.1
44
Table 8: Sources of revenue and the achievement of targets in public good provision
(1) (2) (3) (4)Educational Infant Health Water
VARIABLES enrolment mortality insurance contamination
PANEL A: REDUCED FORM
D(post-cadastral-update)i,t 0.0133 -0.000557 0.0313* 0.0760***[0.0134] [0.0109] [0.0174] [0.0219]
royaltiesoili,00−04 × priceoil
t 0.0292* 0.00155 -0.0380 0.0589[0.0176] [0.00166] [0.0304] [0.0762]
PANEL B: IV
property tax revenuei,t 2.197 -0.0855 4.907* 8.974***[2.093] [1.742] [2.877] [2.823]
natural resource royaltiesi,t 0.175 0.00799 -0.177 0.317[0.118] [0.0164] [0.189] [0.376]
1st stage F-statistic 14.604 14.604 14.604 8.126p-value H0:tax=royalties 0.325 0.957 0.074 0.002
Target met ( % in 2005) 0.15 0.16 0.15 0.18Observations 6,704 6,704 6,704 4,467Number of municipalities 969 969 969 937Dependent variable (dummy for target met) in the header. Money variables in millions of 2004COP per capita. Standardized coefficients for royaltiesoili,00−04 × priceoilt . In panel B, D(post-
cadastral-update) and royaltiesoili,00−04 × priceoilt are used as instruments for property tax re-venue and natural resource royalties, respectively. All regressions include municipality anddepartment-year fixed effects (sample period: 2005-2011, except column 4: 2007-2011). Stan-dard errors clustered two-way by municipality and department-year. *** p<0.01, ** p<0.05, *p<0.1
45
Table 9: Sources of revenue and the achievement of targets in public good provision (cu-mulative royalties)
(1) (2) (3) (4)Educational Infant Health Water
VARIABLES enrolment mortality insurance contamination
PANEL A: REDUCED FORM
D(post-cadastral-update)i,t 0.0133 -0.000528 0.0318* 0.0769***[0.0133] [0.0109] [0.0174] [0.0219]∑t
k=2006 royaltiesoili,00−04 × priceoil
k -0.00966 -0.00132 -0.00219 -0.0735[0.0284] [0.00404] [0.0288] [0.0481]
PANEL B: IV
property tax revenuei,t 2.130 -0.0851 5.085* 9.184***[2.090] [1.743] [2.876] [2.827]
natural resource royalties (cum.)i,t -0.00972 -0.000957 -0.00677 -0.0751**[0.0224] [0.00365] [0.0226] [0.0310]
1st stage F-statistic 11.807 11.807 11.807 7.381p-value H0:tax=royalties 0.306 0.962 0.077 0.001
Target met ( % in 2005) 0.15 0.16 0.15 0.18Observations 6,704 6,704 6,704 4,467Number of municipalities 969 969 969 937Dependent variable (dummy for target met) in the header. Money variables in millions of 2004COP per capita. Standardized coefficients for
∑royaltiesoili,00−04 × priceoilt . In panel B, D(post-
cadastral-update) and∑
royaltiesoili,00−04×priceoilt are used as instruments for property tax revenueand cumulative natural resource royalties, respectively. All regressions include municipality anddepartment-year fixed effects (sample period: 2005-2011, except column 4: 2007-2011). Standarderrors clustered two-way by municipality and department-year. *** p<0.01, ** p<0.05, * p<0.1
46
Table 10: Heterogeneous effect of cadastral updates for oil-royalty recipients
(1) (2) (3) (4)Educational Infant Health Water
VARIABLES enrolment mortality insurance quality
PANEL A: natural log of dependent variable
D(post-cadastral-update)i,t 0.00887*** 0.00301 0.0123 -0.119**[0.00334] [0.00210] [0.00855] [0.0559]
D(post-cadastral-update)i,t × royaltiesoili,00−04 -0.000692 -0.00117 -0.00285 0.0216
[0.00109] [0.000873] [0.00284] [0.0312]
p-value H0: effect for median oil royalties=0 0.009 0.174 0.163 0.040p-value H0: effect for average oil royalties=0 0.021 0.375 0.259 0.126
PANEL B: dummy for target met
D(post-cadastral-update)i,t 0.0136 -0.000637 0.0314* 0.0748***[0.0132] [0.0109] [0.0173] [0.0220]
D(post-cadastral-update)i,t × royaltiesoili,00−04 -0.00943 0.00101 0.00397 -0.00224
[0.0129] [0.00171] [0.00795] [0.0174]
p-value H0: effect for median oil royalties=0 0.365 0.964 0.064 0.001p-value H0: effect for average oil royalties=0 0.792 0.978 0.057 0.008
Observations 6,704 6,704 6,704 4,467Number of municipalities 969 969 969 937
Notes: Dependent variable in the header. royaltiesoili,00−04 is standardized: among oil-royalty
recipients (00-04) the mean is 0.93, median is 0.15. All regressions include municipality anddepartment-year fixed effects (sample period: 2005-2011 except column 4: 2007-2011). Standarderrors clustered two-way by municipality and department-year. *** p<0.01, ** p<0.05, * p<0.1
47
Table 11: Sources of revenue and disciplinary processeses
(1) (2) (3) (4) (5) (6)Mayor Mayor Mayor Top staff Top staff Top staff
VARIABLES prosecuted guilty discharged prosecuted guilty discharged
PANEL A: REDUCED FORM
number of updates -0.000634 -0.00253 -0.00707 0.00512 -0.00247 -0.00883[0.0370] [0.0336] [0.0258] [0.0232] [0.0178] [0.0148]
royaltiesi,2000 × priceoili,t 0.254** 0.220* 0.144** 0.151*** 0.0971 0.109
[0.116] [0.119] [0.0699] [0.0576] [0.0597] [0.0663]
PANEL B: IV
total property tax revenue -0.110 -0.156 -0.267 0.102 -0.110 -0.309[1.147] [1.041] [0.789] [0.710] [0.544] [0.446]
total royalties 0.103** 0.0895* 0.0584** 0.0616*** 0.0394* 0.0439*[0.0511] [0.0516] [0.0260] [0.0185] [0.0221] [0.0227]
Observations 2,888 2,888 2,888 2,888 2,888 2,888Number of codmpio 964 964 964 964 964 964Dependent variable mean 0.19 0.14 0.08 0.06 0.04 0.03Dependent variable in the header. The dependent variables are dummies indicating if a disciplinaryprocess involving the official was opened, whether the official was found guilty and whether the officialwas discharged from office. Columns 1-3 look at mayors, while columns 4-6 look at top executive staff.Panel A shows reduced form results, while panel B shows IV estimates, where predicted royalties(oil royalties in 2000 x oil price index) and the cumulative number of cadastral updates (weighted byshare of cadastre updated) are used as instruments for total royalties and total property tax revenue(Millions of 2004 COP per capita). The first stage F-statistic is 16.35. All regressions include muni-cipality and department-term fixed effects (2004-2007, 2008-2011). Robust standard errors clusteredby municipality and department-term in brackets. *** p<0.01, ** p<0.05, * p<0.1
48
Table 12: Additional effects of oil price shocks
(1) (2) (3) (4)ln Business Murder FARC
VARIABLES population tax rate events
PANEL A: Contemporary shocks
royaltiesoili,00−04 × priceoil
t -0.000188 0.00573 -3.131** 0.594***[0.00191] [0.00485] [1.423] [0.143]
PANEL B: Cumulative shocks∑tk=2006 royaltiesoil
i,00−04 × priceoilk 0.0101*** 0.0149*** -0.558 0.226
[0.00242] [0.00411] [1.929] [0.211]
Observations 6,704 6,704 6,704 5,743Number of municipalities 969 969 969 966Dependent var. mean 30026.40 0.02 33.87 1.37Dependent variable in the header. Money variables in millions of 2004 COP per capita. Allcoefficients are standardized. All regressions include municipality and department-year fixedeffects (sample period: 2005-2011, except column 4: 2005-2010). Standard errors clusteredtwo-way by municipality and department-year. *** p<0.01, ** p<0.05, * p<0.1
49
Table
13:
Th
eeff
ects
ofoi
lpri
cesh
ock
son
mu
nic
ipal
itie
sin
the
top
qu
arti
les
ofth
eoi
lab
un
dan
ced
istr
ibu
tion
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
VA
RIA
BL
ES
Roy
alti
esln
Busi
nes
sM
urd
erFA
RC
Educa
tion
alIn
fant
Hea
lth
Wat
erp
opula
tion
tax
rate
even
tsen
rolm
ent
mor
tality
insu
rance
conta
min
atio
n
PA
NE
LA
:th
ird
quar
tile
-co
nte
mp
orar
yro
yalt
ies
roya
ltie
soili,
00−
04×
pri
ceoilt
0.44
6*-0
.008
600.
0520
-57.
513.
208
0.09
57**
*0.
0204
-0.0
861
-1.0
80[0
.249
][0
.023
5][0
.043
2][4
0.29
][4
.297
][0
.036
6][0
.031
4][0
.186
][1
.510
]
Obse
rvat
ions
6,70
45,
990
5,99
05,
990
5,13
15,
990
5,99
05,
990
3,97
0N
um
ber
ofm
unic
ipal
itie
s96
986
586
586
586
286
586
586
583
3
PA
NE
LB
:to
pquar
tile
-co
nte
mp
orar
yro
yalt
ies
roya
ltie
soili,
00−
04×
pri
ceoilt
0.18
6***
-5.9
3e-0
60.
0048
6-2
.444
0.89
0***
0.00
345
-0.0
0034
8-0
.017
9-0
.241
[0.0
395]
[0.0
0057
5][0
.006
02]
[2.0
42]
[0.3
32]
[0.0
0245
][0
.000
622]
[0.0
151]
[0.1
57]
Obse
rvat
ions
5,97
55,
975
5,97
55,
975
5,11
65,
975
5,97
55,
975
3,95
7N
um
ber
ofm
unic
ipal
itie
s86
286
286
286
285
986
286
286
283
0
PA
NE
LC
:th
ird
quar
tile
-cu
mula
tive
roya
ltie
s
∑ t k=
2006
roya
ltie
soili,
00−
04×
pri
ceoilk
2.28
3***
0.06
63*
0.02
58-8
.452
-2.5
390.
0432
0.01
02-0
.406
*-0
.327
[0.8
16]
[0.0
364]
[0.0
482]
[38.
14]
[5.1
14]
[0.0
706]
[0.0
436]
[0.2
27]
[1.4
90]
Obse
rvat
ions
5,99
05,
990
5,99
05,
990
5,13
15,
990
5,99
05,
990
3,97
0N
um
ber
ofm
unic
ipal
itie
s86
586
586
586
586
286
586
586
583
3
PA
NE
LD
:to
pquar
tile
-cu
mula
tive
roya
ltie
s
∑ t k=
2006
roya
ltie
soili,
00−
04×
pri
ceoilk
1.26
4***
0.00
877*
**0.
0164
***
-1.7
050.
317
0.00
158
-0.0
0107
-0.0
190*
**0.
154
[0.0
876]
[0.0
0214
][0
.003
38]
[2.6
32]
[0.2
33]
[0.0
0297
][0
.002
78]
[0.0
0697
][0
.107
]
Obse
rvat
ions
5,97
55,
975
5,97
55,
975
5,11
65,
975
5,97
55,
975
3,95
7N
um
ber
ofm
unic
ipal
itie
s86
286
286
286
285
986
286
286
283
0N
ote
s:D
epen
den
tva
riab
lein
the
hea
der
(nat
ura
llo
gin
colu
mn
s6-9
).S
am
ple
inp
an
els
Aan
dC
(Ban
dD
)on
lyin
clu
des
mu
nic
ipali
ties
wit
hze
rooil
roya
ltie
sb
etw
een
2000
an
d20
04an
dm
un
icip
alit
ies
inth
eth
ird
(top
)qu
art
ile
of
the
dis
trib
uti
on
of
aver
age
2000-2
004
roya
ltie
s.A
llre
gre
ssio
ns
incl
ud
em
un
icip
ali
tyan
dd
epart
men
t-ye
arfi
xed
effec
ts.
Sta
nd
ard
erro
rscl
ust
ered
two-
way
by
mun
icip
ali
tyan
dd
epart
men
t-ye
ar.
Sam
ple
per
iod
:20
05-2
011,
exce
pt
colu
mn
5(2
005-2
010)
.***
p<
0.0
1,
**
p<
0.0
5,
*p<
0.1
50
Table 14: Sources of revenue and total expenditure
(1) (2) (3) (4)Total Total Total Total
VARIABLES expenditure expenditure expenditure expenditure (cum.)
property tax revenuei,t 1.335 1.779[2.266] [2.014]
natural resource royaltiesi,t−1 0.629 0.646[0.529] [0.544]
natural resource royalties (cum.)i,t 1.179***[0.117]
p-value H0: MPC = 1 0.882 0.482 0.127p-value H0: MPC (taxes) = MPC (royalties) = 1 0.645
Observations 6,704 6,704 6,704 6,704Number of municipalities 969 969 969 9691st stage F-statistic 23.999 22.725 14.047 184.161
Notes: Dependent variable in the header. Money variables in millions of 2004 COP per capita. All regressionsinclude municipality and department-year fixed effects. Instrument for royalties: royaltiesoil
i,00−04×priceoilt−1 (columns
2,3),∑t
k=2006 royaltiesoili,00−04 × priceoil
k (column 4). D(post-cadastral-update) is the instrument for property taxrevenue. MPC: marginal propensity to consume. Sample period: 2005-2011. Standard errors clustered two-way bymunicipality and department-year. *** p<0.01, ** p<0.05, * p<0.1
51
Table
15:
Sou
rces
ofre
venu
e,in
vest
men
tan
dop
erat
ing
exp
ense
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Inve
stm
ent
Inve
stm
ent
Inve
stm
ent
Inve
stm
ent
Op
erat
ing
Op
erat
ing
Op
erat
ing
Op
erat
ing
VA
RIA
BL
ES
Inve
stm
ent
(cu
m.)
exp
ense
sex
pen
ses
exp
ense
sex
pen
ses(
cum
.)
pro
per
tyta
xre
venu
e i,t
1.16
51.
593
0.15
70.
174
[2.1
72]
[1.9
24]
[0.2
52]
[0.2
54]
nat
ura
lre
sou
rce
roya
ltie
s i,t−
10.
608
0.62
30.
0229
0.02
46[0
.516
][0
.530
][0
.020
6][0
.021
8]n
atu
ral
reso
urc
ero
yalt
ies
(cu
m.)i,t
1.12
5***
0.05
35**
*[0
.113
][0
.006
39]
Ob
serv
atio
ns
6,70
46,
704
6,70
46,
704
6,70
46,
704
6,70
46,
704
Nu
mb
erof
mu
nic
ipal
itie
s96
996
996
996
996
996
996
996
91s
tst
age
F-s
tati
stic
23.9
9922
.725
14.0
4718
4.16
123
.999
22.7
2514
.047
184.
161
p-v
alu
eH
0:M
PC
=1
0.93
90.
447
0.27
20.
001
0.00
00.
000
p-v
alu
eH
0:M
PC
taxes
=M
PC
roya
ltie
s0.
590
0.54
5p
-val
ue
H0:
MP
Cta
xes
=M
PC
roya
ltie
s=
10.
647
0.00
0D
epen
den
tva
riab
lein
the
hea
der
.M
oney
vari
able
sin
mil
lion
sof
2004
CO
Pp
erca
pit
a.A
llre
gres
sion
sin
clu
de
mu
nic
ipal
ity
and
dep
artm
ent-
year
fixed
effec
ts.
Inst
rum
ent
for
roya
ltie
s/ca
pit
alre
venu
e:ro
yalt
ieso
ili,
2004×
pri
ceoilt−
1(c
olu
mn
s1-
3,5-
7),∑ ro
yalt
ieso
ili,
2004×
pri
ceoilt
(col
um
ns
4,8)
.D
(pos
t-ca
das
tral
-up
dat
e)is
the
inst
rum
ent
for
pro
per
tyta
xre
venu
e.S
amp
lep
erio
d:
2005
-201
1.S
tan
dar
der
rors
clu
ster
edtw
o-w
ayby
mu
nic
ipal
ity
and
dep
artm
ent-
year
.**
*p<
0.01
,**
p<
0.05
,*
p<
0.1
52
Table 16: Sources of revenue and educational infrastructure
(1) (2) (3) (4)Schools Schools Teaching area Teaching area
VARIABLES per 10,000 inh. per 10,000 inh.
PANEL A: REDUCED FORM
D(post-cadastral-update)i,t 0.897** 0.196 978.2 56.04[0.440] [0.146] [1,134] [236.1]∑t
k=2006 royaltiesoili,00−04 × priceoil
k 0.206 0.0479 637.9 657.1[0.164] [0.142] [1,008] [731.3]
PANEL B: IV
property tax revenuei,t 143.7* 31.38 112,653 7,231[74.43] [25.18] [132,842] [26,888]
natural resource royalties (cum.)i,t 0.0195 0.00653 374.7 548.1[0.149] [0.116] [815.9] [575.3]
1st stage F-statistic 11.807 11.807 5.503 5.503p-value H0:tax=royalties 0.054 0.213 0.399 0.804
Observations 6,704 6,704 3,882 3,882Number of municipalities 969 969 871 871Dependent variable mean 47.69 27.93 10,010.2 4,852.09Dependent variable in the header. Money variables in millions of 2004 COP per capita. Standardized coefficientsfor
∑royaltiesoili,2004 × priceoilt . In panel B, D(post-cadastral-update) and
∑royaltiesoili,2004 × priceoilt are used as
instruments for property tax revenue and cumulative natural resource royalties, respectively. All regressions includemunicipality and department-year fixed effects. Sample period: 2005-2011 (columns 1 and 2), 2006-2011 (columns3 and 4). Standard errors clustered two-way by municipality and department-year. *** p<0.01, ** p<0.05, * p<0.1
53
ONLINE APPENDIX
Appendix A Theoretical Appendix
A.1 Set-up of the model and equilibrium
This is a two-period model in which a citizen/voter obtains utility from private consum-
ption of her disposable income and also from consumption of a public good. At the end of
the first period an election between the incumbent and a random opponent takes place. The
incumbent as well as his opponent are drawn from a pool of potential politicians, each en-
dowed with some level of ability θi > 0. The ability of all politicians is unknown to everyone
but there is a common prior that is normally distributed with mean m and precision h.
The citizen receives a constant income yt = y each period. She pays a tax on the fraction
of her income η ∈ (0, 1) at an exogenous rate τ ∈ (0, 1). The citizen’s private consumption is
equal to her disposable income: ct = (1− τη)yt. Her utility function is Ut = U(ct, gt), where
gt is the amount of the public good that is supplied that period. U(·) is increasing in both
of its arguments.
Government revenue (Rt) is equal to tax revenue (amounting to τηyt) plus revenue from
an external source (Tt) such as royalties from the extraction of natural resources or transfers
from another level of government. I assume that operational expenditures eat up the constant
share of revenue 1− µ, so the amount of revenue available for public good provision is µRt,
µ ∈ (0, 1).
The amount of public good provided by the mayor with ability level θ at time t ∈ {1, 2}is given by the function
gt = θ + µRt + et (5)
where Rt = τηyt + Tt and et ≥ 0 is the amount of effort put in by the mayor, which is
unobservable to the citizen.29 The cost of effort borne by the mayor is given by the increasing
and strictly convex function γC(e), γ > 0. The mayor also gets a benefit E > 0 from being in
power each period, which includes financial rewards and “ego rents”. Total per-period utility
for the mayor is then E − γC(e).
At the end of the first period the citizen observes the amount of public good provided.
She also receives a noisy signal (Rt) on the total amount of revenue (Rt). I assume that the
signal is equal to the actual revenue minus a mean-zero normally distributed noise term:
Rt = Rt + εRt where εRt ∼ N [0, 1/hεR] Based on this information and a conjecture on effort
she updates her beliefs on the incumbent’s ability. She then votes for the candidate of her
29See Dewatripont et al. (1999) for a discussion of more general versions of this type of model.
54
liking.
Before making additional assumptions about the link between the sources of revenue and
the noisy signal that the citizen receives, I summarize the timing of the game (I will drop
the time subscripts for everything that is not changing over time):
1. The incumbent (with ability θI unknown to all) receives revenue R = T + τηy and
chooses the amount of effort e1.
2. A quantity of public goods g1 is provided according to equation 5.
3. The citizen observes g1 and receives the noisy signal R. She uses this information to
update her beliefs on the incumbent’s ability: θI .
4. The citizen votes either for the incumbent or for a random opponent with the same
prior ability (m).
5. The winner of the election chooses e2 and this determines g2.
We can find the PBE in pure strategies using backwards induction. The winning candidate
(with ability θ2) solves the following problem in the second period:
maxe2≥0
E − γC(e2)
Since γC(·) is an increasing function, the second-period mayor will set e?2 = 0 and will get
utility E that period. Therefore, the amount of public good provided will be g?2(θ2) = θ2+µR,
which is an increasing function of the ability of the second-period mayor. If the voter chooses
a candidate with believed ability θ, her second period utility is U(c2, g?2(θ)) = U(1− τηy, θ+
µR). Since U(·) is increasing in g2, the incumbent will be re-elected only if the voter believes
him to have higher ability than the opponent. That is, if θI ≥ m.
The citizen updates her beliefs on the incumbent’s ability based on the amount of first-
period public goods (g1), her conjecture on the level of effort put in by the incumbent
(e1), and the noisy signal on revenue (R). The problem that the citizen faces is that the
discrepancy between the observed amount of public goods and the amount she expected,
which I will label Z1 ≡ g1−µR− e1, is equal to the incumbent’s ability (θI) minus the noise
term’s impact on expected public goods (µεRt ). Given that θI and µεRt are two independent
normally distributed random variables, the solution to the signal extraction problem is:
θI = E[θI |Z1] =mh+ Z1hRh+ hR
(6)
55
This expression says that the posterior belief on the incumbent’s ability is a weighted average
of the prior m and the discrepancy Z1, where the respective weights are given by the precision
of the prior (h) and of the noise term (hR = hεR/µ2). As the signal gets noisier (hR → 0), it
becomes less informative and the posterior gets closer to the prior. Similarly, as the signal
gets more precise (hR → ∞) it perfectly reveals the incumbent’s ability and full updating
occurs (all the discrepancy is attributed to ability). Equation 6 implies that the re-election
condition simplifies to Z1 ≥ m.
From the incumbent’s perspective, his probability of re-election is:
pI(e1) = prob(g1 − µR− e1 ≥ m)
= prob(θ + µR + e1 − µR− µεRt − e1 ≥ m)
= prob(θ − µεRt ≥ m+ e1 − e1)
= prob(Z1 ≥ m+ e1 − e1)
= 1− ΦZ(m+ e1 − e1)
where Φ is the cumulative distribution function of the normally distributed Z1, which has
mean m and precision hZ ≡ h·hRh+hR
. The expression above tells us that the incumbent can
increase his probability of re-election by increasing the amount of effort (e1) relative to the
voter’s conjecture (e1). Therefore, the problem solved by the incumbent in period 1 is:
maxe1≥0
E − γC(e1) + (1− ΦZ(m+ e1 − e1)) βE
Assuming an interior solution, the first-order condition of this problem is:
γC ′(e?1) = φz(m+ e1 − e?1)βE
where φz is the probability density function of Z1. Given that in equilibrium e1 = e?1, the
first-order condition characterizing optimal first-period effort simplifies to:30
γC ′(e?1) =βE√2π/hZ
(7)
As equation 7 shows, e?1 does not depend on revenue. Hence, a $1 increase from either
source has a mechanical and homogeneous effect of size µ on public good provision.31 Equa-
30The equilibrium re-election probability is thus 1 − Φz(m) = 1/2 since m is the mean of the normallydistributed Z1.
31In a modified version of the model, in which the incumbent’s choice is over private rents rather thaneffort, revenue does have a positive effect (the electoral cost of a fixed amount of rents decreases as revenueincreases), but this effect is still homogeneous across sources. See Persson and Tabellini (2000), Alesina and
56
tion 7 also shows that e?1 is an increasing function of hZ (since C(e) is strictly convex), which
is itself an increasing function of the precision of the revenue signal hR. Hence, as the signal
becomes more precise, the voter becomes more attentive to public good provision and the
incumbent provides more effort in equilibrium.
A.2 Taxes as a source of information on public revenue
Having solved the model, I now examine two extensions that link the source of revenue to
the precision of the revenue signal and yield predictions of a heterogeneous effect of revenue
from different sources on incumbent effort and public good provision. I start by assuming
that the share of exogenous revenue amplifies the noise in the voter’s signal:
Assumption 1. Rt = Rt − εRt where εRt = εt ×(
TT+τηy
) 12
and εt ∼ N [0, 1/hε]
Hence, the precision of the revenue signal is hR =(T+τηyT
)hεµ2
. This assumption captures
the idea that citizens are better informed about changes in tax revenue than about changes
in external revenue. Now, as tax revenue increases, the signal becomes more precise and the
citizen becomes more attentive to the amount of public goods provided in her assessment of
the incumbent’s quality. This in turn makes it optimal for the incumbent to increase effort
in order to influence the election in his favour. By the same logic, an increase in exogenous
revenue makes the revenue signal noisier and the citizen less responsive, so the incumbent
reduces effort.
As before, the functional form of the production function for public goods implies that
revenue from any source has a mechanical effect on public good provision. However, the total
or net effect of a revenue increase depends also on the indirect effect through incumbent effort.
Under assumption 1, an increase in tax revenue has a larger effect on public good provision
than an equivalent increase in external revenue due to the opposite indirect effect through
incumbent effort. The following proposition formalizes this result.
Proposition 1. Under assumption 1, the equilibrium first-period effort of the incumbent is
increasing in tax revenue and decreasing in external revenue. Hence, public good provision in
the first period increases by more than the mechanical revenue effect when there is an increase
in tax revenue and by less than the mechanical revenue effect when there is an increase in
external revenue.
Tabellini (2007), Brollo et al. (2013) or Matsen et al. (2015) for examples. If the incumbent is unconstrainedon the amount he can appropriate, extra revenue has the additional effect of increasing the value of stayingin power. This mechanism is at play in the model of the resource curse in Robinson et al. (2006). Still, thisdoes not give rise to any heterogeneity across sources.
57
Proof. Using the implicit function theorem and noting that C(·) is a strictly convex function,
when we differentiate (7) with respect to τηy we obtain:
γ∂2C
∂e?12
∂e?1∂τηy
=∂φZ∂hZ
∂hZ∂τηy
βE (8)
From φZ =√hZ/2π we can see that ∂φZ/∂hZ > 0. Using the definitions of hZ and hR
we find that∂hZ∂τηy
=(µh)2hεT
(hε(T + τηy) + µ2hT )2 > 0
Since all other terms on both sides of equation 8 are positive,∂e?1∂τηy
> 0. Hence, the overall
effect of a tax revenue increase on first-period public good provision, based on equation 1, is
given by∂g?1∂τηy
= µ+∂e?1∂τηy
> µ
where µ is the mechanical revenue effect.
Using again implicit differentiation on equation (7) but with respect to T we get
γ∂2C
∂e?12
∂e?1∂T
=∂φZ∂hZ
∂hZ∂T
βE (9)
The argument works the same as in the case of taxes, except that now
∂hZ∂T
=−(µh)2hετηy
(hε(T + τηy) + µ2hT )2 < 0
Since all the other terms on both sides of equation 9 are positive,∂e?1∂T
< 0. Therefore,
the overall effect of an increase in exogenous revenue on first-period public good provision is
given by∂g?1∂T
= µ+∂e?1∂T
< µ <∂g?1∂τηy
where again µ is the mechanical revenue effect.
A.3 Taxes as an incentive for information acquisition on public
revenue
I now substitute Assumption 1 with the following three assumptions:
Assumption 2. Rt = Rt + εRt , where εRt ∼ N [0, 1/hεR]
58
Assumption 3. Ut = U(ct +αgt) where U(·) is a strictly concave function and α ∈ (0, 1/µ)
Assumption 4. At the start of the game, the voter can choose how much effort (f1 ≥ 0) to
spend on the improvement of the revenue signal. Effort increases the precision of the revenue
signal according to the linear function hεR = λf1, λ > 0, but has a cost given by the strictly
convex function K(f1)
Under assumptions 2-4 the model is essentially unchanged: the incumbent’s first-period
effort is still determined by (7) and is increasing on the precision of the revenue signal.
If we substitute the public good production function and the voter’s budget constraint
into her first-period utility function we can see that the problem the voter solves is
maxf1≥0
U1 = U(c1 + αg1)−K(f1)
= U [(1− τη)y + α (θI + µ(τηy + T ) + e?1(hZ))]−K(f1)
= U [y + (αµ− 1)τηy + αµT + αθI + αe?1(hZ)]−K(f1)
Assuming an interior solution, the optimal amount of voter effort is implicitly defined by
the following first-order condition:
U ′[y + (αµ− 1)τηy + αµT + αθI + αe?1(hZ)]α∂e?1∂hZ
∂hZ∂hεR
λ = K ′(f ?1 ) (10)
Just like before, the voter is more responsive to public good provision in her assessment
of the incumbent’s quality the better she is informed about revenue. In turn, the incumbent
puts in more effort as the voter becomes more responsive. Under the new assumptions, what
sets this mechanism in motion is information acquisition by the voter, which depends on the
marginal utility of public goods. When tax revenue increases, private consumption mechani-
cally decreases. Although public good provision also increases due to the mechanical revenue
effect, assumption 3 ensures that the marginal utility of the public good increases as well,
which raises the benefit the voter gets from additional incumbent effort.32 External revenue,
on the other hand, has a negative effect on the marginal utility of the public good due to the
positive mechanical revenue effect and the fact that it does not affect the voter’s disposable
income. Hence, extra taxation provides an incentive for more information acquisition while
the opposite holds true for external revenue. The following proposition formalizes this result:
32Assumption 3 implies that taxation for the purpose of public good provision is inefficient in a settingwhere overhead costs are relatively large. This is consistent with the findings of Pritchett and Aiyar (2014),who report that the median cost of one pupil in public elementary school in India is twice as high as in aprivate school, but educational achievement is much lower.
59
Proposition 2. Under Assumptions 2-4, the equilibrium first-period effort of the voter and
the incumbent are increasing in tax revenue and decreasing in external revenue. Hence, public
good provision in the first period increases by more than the mechanical revenue effect when
there is an increase in tax revenue and by less than the mechanical revenue effect when there
is an increase in external revenue.
Proof. Using the implicit function theorem we can differentiate both sides of (10) with
respect to τηy to obtain:
αλ∂e?1∂hZ
∂hZ∂hεR
U ′′(·)(
(αµ− 1) + α∂e?1∂hZ
∂hZ∂hεR
λ∂f ?1∂τηy
)= K ′′(f ?1 )
∂f ?1∂τηy
⇒ αλ∂e?1∂hZ
∂hZ∂hεR
U ′′(·)(αµ− 1) =∂f ?1∂τηy
(K ′′(f ?1 )− α2λ2
(∂e?1∂hZ
∂hZ∂hεR
)2
U ′′(·)
)(11)
Since U(·) is a strictly concave function whileK(f1) is strictly convex, the LHS in equation
11 is positive and the term in brackets on the right is also positive. Hence, ∂f ?1 /∂τηy > 0.
This implies, from equation 1, that the overall effect of an increase in tax revenue on public
good provision is given by
∂g?1∂τηy
= µ+∂e?1∂hZ
∂hZ∂hεR
λ∂f ?1∂τηy
> µ
Using implicit differentiation on equation 10 but with respect to exogenous revenue (T )
yields:
αλ∂e?1∂hZ
∂hZ∂hεR
U ′′(·)αµ =∂f ?1∂T
(K ′′(f ?1 )− α2λ2
(∂e?1∂hZ
∂hZ∂hεR
)2
U ′′(·)
)(12)
Now the LHS in equation 12 is negative, while the term in brackets on the right remains
positive. Hence, ∂f ?1 /∂T < 0. Using again equation 5, we can see that the net effect of an
increase in exogenous revenue on first-period public good provision is
∂g?1∂T
= µ+∂e?1∂hZ
∂hZ∂hεR
λ∂f ?1∂T
< µ <∂g?1∂τηy
60